Minerals Engineering 74 (2015) 4150Contents lists available at ScienceDirect

Minerals Engineering

journal homepage: www.elsevier .com/locate /minengOptimization of a fully air-swept dry grinding cement raw meal ball millclosed circuit capacity with the aid of simulationhttp://dx.doi.org/10.1016/j.mineng.2015.01.0060892-6875/ 2015 Elsevier Ltd. All rights reserved.

Tel.: +90 252 2111938; fax: +90 252 2111912.E-mail addresses: ogenc@mu.edu.tr, omurdeng@gmail.com. Gen Mugla Stk Koman University, Faculty of Engineering, Dept. of Mining Engineering, Ktekli, Mugla 48000, Turkey

a r t i c l e i n f oArticle history:Received 19 August 2014Accepted 9 January 2015

Keywords:GrindingClassificationModellingSimulationOptimizationa b s t r a c t

Production capacity of a fully air-swept industrial scale two-compartment KHD Humboldt Wedag

cement ball mill was optimized with the aid of simulation. It was proposed to operate the mill as a singlecompartment by eliminating the pre-drying compartment. In this respect, grinding performance of theair-swept ball mill was evaluated and modelled as a perfectly mixed single tank using the perfect mixingball mill modelling approach (Whiten, 1974). Static separator was modelled by efficiency curve model(Whiten, 1966). The empirical breakage function required in the estimation of average specific breakagerates was measured by drop-weight technique. The full scale model parameters were used to simulatethe raw meal mill grinding circuit with the aid of JKSimMet Steady State Mineral Processing Simulator.Simulation results indicated 23% production capacity increase in cement throughput in case the pre-drying compartment was used in grinding.

2015 Elsevier Ltd. All rights reserved.1. Introduction

Air-swept raw meal ball mills introduced by the cement millmanufacturers F.L.Smidth

(Smidth, 2002), Polysius

(Polysius,

2002) and KHD Humboldt Wedag are the most commonly usedones. KHD Humboldt Wedag manufactured fully air-swept rawmeal mills which have two compartments used for drying andgrinding processes. In these mills drying and grinding are per-formed in a single mill as similar to the Polysius

fully air-swept

mill (Polysius, 2002). First compartment is used as a pre-dryingcompartment where it is equipped with lifters and operated with-out grinding media in order to increase the drying efficiency. Insuch systems, kiln discharge gases are used as a drying air. Dryingcompartment consumes more energy as compared to the othersystems due to the high level of moisture in the feed. In air-sweptmills circulating load is carried pneumatically. Thus, the energyconsumption for a fully air-swept grinding circuit is higher byapproximately 1012% as compared to the grinding circuit withbucket elevator (Duda, 1985). Modelling of fully air-swept ballmills used in the cement industry were studied with differentapproaches in the literature (Austin et al., 1975, 1984;Viswanathan, 1986; Viswanathan and Narang, 1988;Viswanathan and Reddy, 1992; Zhang et al., 1988; Zhang, 1992;Ergin, 1993; Apling and Ergin, 1994; Benzer, 2004). Grinding modelparameters are similar except of the material transport function inthe related models. The population balance model requires resi-dence time which is difficult to determine for the full-scale mill.Value of residence time distribution is determined experimentally.Perfect mixing model (Whiten, 1974) simplifies the discharge(transport) function by assuming a particle size dependent dis-charge rate function. The discharge of any particle fraction fromthe mill can be calculated on the basis of the mass of size fractionin the mill hold-up and mass flow rate of that particle fraction outof the mill as product. Perfect mixing model does not constitutemany grinding parameters which needs to be scaled up. The modelcould be used directly to predict the performance of full-scalemills. The relation between particle size and discharge rate depen-dent breakage rate parameter which was defined as a ratio ofbreakage rate to discharge rate function was established to mea-sure the ball milling performance based on perfect mixing model-ling approach by Zhang (1992), Benzer (2000) and Hashim (2003).

Breakage function and breakage rate parameters are deter-mined by laboratory experiments in Austins approach (Austinet al., 1984) and the resulting mathematical equations relatingthe breakage function and breakage rate to particle size constitutemany parameters. Thus, more than one parameter set could beproduced in the solution of these equations each of which definedifferent breakage rate-particle size relationships. For this reason,it is difficult to relate the effects of operating variables of ball millson specific breakage rates. Design and operational parameterswere studied on laboratory scale mills which need to be

http://crossmark.crossref.org/dialog/?doi=10.1016/j.mineng.2015.01.006&domain=pdfhttp://dx.doi.org/10.1016/j.mineng.2015.01.006mailto:ogenc@mu.edu.trmailto:omurdeng@gmail.comhttp://dx.doi.org/10.1016/j.mineng.2015.01.006http://www.sciencedirect.com/science/journal/08926875http://www.elsevier.com/locate/mineng

Nomenclature

i particle size fraction ij particle size fraction jfi mass flowrate of mill feed (ton/hour)pi mass flowrate of mill discharge (ton/hour)ri specific breakage rate of size fraction i (h1)di specific discharge rate of size fraction i (h1)di normalized discharge rate of size fraction ia single column step triangular breakage function matrixsi mass of size fraction i (ton)Q volumetric feed rate (m3/h)D mill diameter (m)L mill length (m)

r/d ratio of breakage rate to normalized discharge rateEoa fraction of feed reporting to overflowC fraction undergoing real classification (1-bypass frac-

tion)B reduced efficiency curve fish hook parameterd50c size of a particle in feed which has equal probability of

going to underflow or overflow (cut size)b model parameter to preserve the definition of d50cd particle sizex ratio of di to d50ca reduced efficiency curve sharpness parameter

42 . Gen /Minerals Engineering 74 (2015) 4150scaled-up. There had been a few attempts to relate their modelwith air flow through the mill, feed rate, feed size distribution,material filling and ball filling (Viswanathan, 1986; Zhang, 1992).Air swept ball mill model proposed by Austin et al. (1975) was val-idated by Apling and Ergin (1994) using the industrial scale datafrom a cement grinding circuit.

In this study, production capacity of a fully air-swept dry grind-ing raw meal ball mill circuit was evaluated by modelling the millusing the perfect mixing modelling approach (Whiten, 1972). Sta-tic separator in the circuit was modelled by efficiency curve model(Whiten, 1966). JKSimMet Steady State Mineral Processing Simula-tor was used in the simulation stage. Simulation results indicated23% capacity increase in cement throughput at the steady statecondition. However, the static separator is expected to operatewith the maximum tonnage that can be handled.

2. Methods

2.1. Sampling survey

The simplified process flowsheet of the sampled circuit with thesampling points is given in Fig. 1. Air-swept ball mill is operating inclosed circuit with a static separator. The static fines are collectedin product cyclones where the separation of particles from the airis performed. Product of electrofilter is combined with the cycloneproducts to form final cement. Design specifications of the fullyair-swept ball mill and static separator are given in Table 1. Designball size distribution applied in the ball mill is given in Table 2.

Steady state condition of the circuit was verified by examiningthe variations in the values of operational variables of the ball milland the static separator in the process control room system. Sam-pling was started when the steady state condition was achieved.Representative amount of samples were collected from the shownsampling points in Fig. 1. Samples from the raw meal feed werecollected for the determination of moisture content of the mill feedmaterials. Values of the operational variables were recorded inevery 5 min from the process control system to be used in the cir-cuit performance assessment during sampling Control roomrecordings and related standard deviation values at the steadystate condition are tabulated in Table 3.

2.2. Experimental

Samples were prepared by using a riffler for dry sieving fromthe top size down to 150 lm. Sub-sieve sample (150 lm) wassized in wet mode in a SYMPATHEC laser diffractometer. Drysized material (+150 lm) and wet sized sub-sieve sample(150 lm) were combined to define the full size distribution fromthe top size down to 1.8 lm. Raw meal materials were dried atapproximately 100 C before sizing in order to carry out an efficientscreening operation. Calculated moisture contents and dry flow-rates of mill feed materials are given in Table 4.

3. Results and discussions

3.1. Mass balancing

Measured particle size distributions and operational tonnageflowrates were used to perform mass balance calculations aroundthe circuit with the aid of mass balance module of the JKSimMetsimulator to calculate the best fit estimates of the size distributionsand tonnage flowrates. Mass balanced flowrates and calculatedfineness as 0.045 mm passing % are given in Table 5. Circulatingload ratio was defined as the ratio of static separator reject tonnageto static separator fine tonnage and calculated as 75.34%. Theresults of mass balance calculations were checked out by plottingthe experimental and calculated particle size distributions(Fig. 2). Experimental versus mass balanced particle size distribu-tions were found to be fitted satisfactorily which indicated that,sampling was successful and the data could be used for modellingpurpose. Experimental size distributions of final cement cyclonecollectors were presented in Fig. 3. Particle size distributions indi-cated no segregation in the cyclones verifying the sufficient level ofair flow and balanced air distribution within the cyclones.

3.2. Mill inside sampling and granulometry

The circuit was crash-stopped to collect samples from inside ofthe mill after completing sampling of the circuit streams. A view ofmill inside at the crash-stop condition is given in Fig. 4. Averagematerial height above the ball surface level (18 cm) and free heightof the mill (2.27 m) were measured to be used in mill powder load(hold-up) calculation ahead of collecting the samples along thelong axis of the mill at the crash-stop condition. Mill filling was cal-culated to be 32% using the mentioned geometrical measurements.Photograph of the lifter bar design in the drying compartment ispresented in Fig. 5. Considerable abrasion and damage on lifterswere recognized. Whole length of the grinding compartment waslined with classifying liners. Classifying liner configuration is pre-sented in Fig. 4.

Sample collection dips were formed by digging out the millcharge (mill powder + balls) approximately 40 cm below thecharge level. Samples were collected along the long axis of the milltowards the end of the discharge grate in order to demonstrate thesize reduction performance using the inside mill size distributions(granulometry). Samples were collected by one meter up to thesixth meter of the grinding length whereas by half meter at the restof the mill length. Mill inlet and outlet temperatures were recorded

Fig. 1. Simplified flowsheet of a raw meal grinding-classification circuit. Streams/sampled: (1) iron ore bunker belt; (2) clay bunker belt; (3) limestone bunker belt; (4) totalfresh feed; (7) static separator reject (coarse); (9a) product cyclone-1 underflow; (9b) product cyclone-2 underflow; (10) product cyclone combined; (12) electrofilter return;(13) dust from cooler. Streams/not sampled: (5) mill feed; (6) mill discharge; (8) static separator fine; (11) cyclone dust.

Table 1Design specifications for air-swept raw meal ball mill and static separator.

Raw meal ball millDiameter (m) 3.8Drying compartment length (m) 2.935Grinding compartment length (m) 6.935Mill power (kW) 1600Mill rotational speed (rev/min) 15Critical speed % 69Ball filling % 27Discharge diaphragm middle grate aperture size (cm) 8 8Static separatorSeparator diameter (m) 5.2

Table 2Design ball size distribution of the grinding compartment.

Ball size (mm) Weight (kg) Weight % Cumulative weight %

80 2853 4 100.0070 17,552 22 96.4560 18,594 23 74.5950 17,040 21 51.4440 17,442 22 30.2230 6827 9 8.50

Total 80,308 100

Table 3Control room recordings during the sampling survey.

Operational variables Value Standard deviation

Limestone (t/h) 65 1.38Clay (t/h) 26 1.37Iron ore (t/h) 1.72 0.13Total fresh feed wet flowrate (t/h) 92.72 1.55Static separator reject (t/h) 64 14.40Ball mill filling % 83 1.44Ball mill inlet temperature (C) 325 2.96Ball mill discharge temperature (C) 93 3.31Ball mill inlet pressure (mmSS)a 25 4.04Ball mill discharge pressure (mmSS) 360 22.46Ball mill ventilation pressure (mmSS) 767 27.79Static separator pressure difference (mmSS) 335 19.49Ball mill (Amper) 120 0.00Ball mill elevator (Amper) 24 0.00Ball mill motor (kW) 1240 7.90Mill specific energy consumption (kW h/t) 14.57 0.26Kiln capacity (t/h) 71

a Millimeters of water column.

. Gen /Minerals Engineering 74 (2015) 4150 43as 325 C and 93 C respectively at the crash-stop condition. Themill was cooled down for 67 h before inside mill sampling byopening the mill inlet. Air flow through the was not allowed as fineparticles will discharge from the mill.

It should be mentioned that, it is crucial to collect representa-tive samples in any sampling operation. The technique used in thisstudy provided collecting representative inside mill samples at theregarding sample collection dip. Collection of material and ballsamples just above the charge surface (which is common in suchsampling procedures) will not give statistically representativeresults for the evaluation of ball charge load and distribution whichaffects the size reduction performance of the mill. Sample amountcollected at each sample collection dip were tabulated in Table 6.Mill length given in Table 6 refers to the measured length at thesampling condition. Particle size distribution at the mill inlet wasfound to be coarser than that of the following sampling dipsexcluded of the particle size distribution of the sample collectedfrom the first meter of the mill length. This condition could berelated to the difficulty of digging of the sample collection dip atthe first meter due to the existing coarse balls such as 90 mmand 80 mmwhich could have affected the quality of sampling. Par-ticle size distribution of the mill inlet was found to be finer thanthe first meter sample as shown in Fig. 6. This condition could bedue to the accumulation of static separator reject material at themill inlet which affected the particle size distribution at thecrash-stop condition. Particle size distributions of the inside millsamples and the mass balanced mill feed and discharge size distri-butions are presented on loglog scale in Fig. 6. Particle size distri-bution of the mill hold-up (mill load) was assumed to be calculatedusing the average size distribution of the inside mill samples whichis denoted by the average mill content size distribution in Fig. 6.

The mill modelling approach was to use average mill hold-upparticle size distribution when calibrating the model parametersof perfect mixing model proposed by Whiten (1974). Inside millparticle size distributions (Fig. 6) indicated a consistent size reduc-tion towards the mill discharge end such that, particle size distri-bution of the samples became finer towards the discharge grate.

Mill inside fineness curve established using the 0.045 mmcumulative passing % size is given in Fig. 7. Amount of fine materialproduction in the first meter decreased. However, fine materialproduction increased in the following two meters. No more

Table 4Moisture contents of mill feed and calculated dry flowrates.

Raw meals Moisture % Measured wetflowrate (t/h)

Dry flowrate(t/h)

Limestone 2.08 65 63.65Clay 22.64 26 20.11Iron ore 4.12 1.72 1.65Total raw meal 6.64 92.72 85.41

44 . Gen /Minerals Engineering 74 (2015) 4150considerable size reduction was achieved at the rest of the milllength which could be due to a series of operational factors asgiven below:

probable increase in amount of fine material due to the low airflow rate, such that, less fines extracted from the mill,

increase in mill inside temperature which could lead to cush-ioning effect as explained by Austin et al. (1984). Coating of ballsurface with material is expected to have an adverse effect ongrinding performance of the grinding media thus will result inlower specific breakage rate,

probable agglomeration of fine particles inside the mill whichcould have decreased the transportation (discharge) rate ofparticles through the mill. This claim could be supported byTable 5Mass balanced flowrates and fineness as 0.045 mm passing %.

Stream No Stream identification Sample am

1 Iron ore bunker belt 64.862 Clay bunker belt 47.363 Limestone bunker belt 55.774 Total fresh feed 5 Mill feed 6 Mill discharge (static separator feed) 7 Static separator reject (coarse) 5.368 Static separator fine 9a Product cyclone-1 underflow 2.569b Product cyclone-2 underflow 2.6010 Product cyclone combined 2.6511 Cyclone dust 12 Electrofilter return 3.2413 Dust from cooler 4.5514 Final cement 2.42

Fig. 2. Agreement between experimental and mass bthe work of (Kolacz, 1999). Effect of air flowrate on the dis-charge rate of material in an air swept ball mill was studiedby Kolacz (1999). It was concluded that, transportation of mate-rial through the mill by air sweeping becomes more difficult ifthe mill content is finer which is due to the agglomeration ofvery fine particles falling back into the mill bed,

material coating observed at the discharge grate couldhave affected the fine material accumulation amount in themill and decreased the grinding performance of the grindingmedia.

Particle size distribution of the mill discharge estimated bymass balance calculations was found to be finer than that of thesample collected at the mill discharge end which correspondedto the sample at the seven point fourth meter of the grindinglength. This condition is expected under sufficient screening effectof the discharge diaphragm (Fig. 6). Screening effect wasexplained as the rejection of coarse particles to the last meter ofthe compartment length after screening at the diaphragm and dis-cussed in the literature (Benzer, 2000; Gen, 2008; Gen andBenzer, 2009) for intermediate and discharge diaphragms of over-flow (gravity discharge) type multi-compartment cement grindingball mills.ount (kg) Calculated flowrate (t/h) 0.045 mm passing %

1.65 3.5720.07 2.7963.41 1.5885.13 1.91

149.26 8.00149.26 52.3264.14 16.2985.13 79.58 80.00 78.80

82.73 78.102.40 100.004.97 100.002.57 100.00

87.70 80.71

alanced size distributions of the circuit streams.

Fig. 3. Experimental size distributions of final cement cyclone collector productsand dust from cooler upstream.

Classifying liners

Grinding compartment

Fig. 4. Photographs of mill inside and classifying liners in the grindingcompartment.

Fig. 5. A view of lifter bar liners in the drying compartment.

Table 6Mill inside sample amounts.

Length (m) Sample (kg)

Mill inlet 10.421 6.602 7.963 5.794 3.915 3.436 3.836.7 3.727.4 6.30

Fig. 6. Axial mill inside particle size distributions towards the mill discharge end.

0

20

40

60

80

100

0 1 2 3 4 5 6 7 8

0.04

5mm

cum

ulat

ive

pass

ing

%

Grinding compartment length (m)

Fig. 7. Fineness variation along the grinding compartment length.

. Gen /Minerals Engineering 74 (2015) 4150 45Mill powder was expected to discharge through the middlegrate of the discharge diaphragm as the grate opening was wideenough (8 8 cm) to allow transportation of finely ground rawmeal powder by only air sweeping in the studied mill. Mill insidesize distributions demonstrated consistent size reduction. Particlesize distributions of the sample at the seventh meter of the grind-ing compartment length was found to be considerably coarser thanthat of the mill discharge (Fig. 6). Both size distributions should beclosely similar under the effective air flowrate operationalconditions.

Another observation was the existence of coarse particle accu-mulation in the mill. Certain amount of coarse particle accumula-tion within the size range of 25 + 19 mm, 19 + 13.2 mm,13.2 + 9.5 mm was observed at the fourth meter of the milllength. Such operational inefficiencies were attributed to the hard-ness of these particles, material coating at the discharge diaphragmand low air flowrate condition as the operational air flowrate at themill outlet was recorded to be 25.9 m/s. Typical range for the airflowrate at the mill outlet is 24.435.1 m/s for air swept ball mills(Duda, 1985). Recorded low air flowrate at the mill outlet couldhave decreased the grinding capacity due to the transportation offine material through the mill. On the otherhand, air flowratethrough the mill was calculated as 5.02 m/s using the measuredmill filling (32%) at the sampling condition. This figure was foundto be higher than the typical air flowrate range suggested for air-swept mills which is 34 m/s inside the mill as given by Duda(1985).

Fig. 8. Weighted measured ball size along the mill length.

46 . Gen /Minerals Engineering 74 (2015) 41503.3. Ball size classification

In order to assess the classifying performance of the mill liners,ball samples were collected during the inside mill powder sam-pling by screening out the balls over a screen with 25 25 mmaperture size in order to separate the raw meal powder and thegrinding media at the sampling dips. Ball samples were collectedevery meter, up to the fourth meter of the mill. The sampling pro-cedure was to collect some amount of mixture of raw meal powderand balls and then screening. Balls were retained on the screen andcollected in a sampling bag to be weighted and sized to determinethe ball size distribution along the mill length. On the otherhand,raw meal powder which was the screen undersize was collectedin another sampling bag. Collected ball sample mass along the milllength was tabulated in Table 7. It should be mentioned that, thepresented values are not representative of the whole ball load atthe sampling dip. However, the results clearly indicated the ballsize classification along the long axis of the mill. Ball size distribu-tion was found to get finer towards the mill discharge end, exceptfor the sample collected at the second meter of the compartmentwhich indicated true ball size classification. The concept of trueball size classification was discussed for cement grinding multi-compartment ball mills by Gen et al. (2008). This condition showsthe affect of classifying liners. Weighted average ball size was cal-culated using the collected ball samples at each sampling locationwhich demonstrated the true ball size classification along the con-sidered mill length and given in Fig. 8.Fig. 9. Normalized single particle breakage functions (replotted after Gen et al.,2008).3.4. Material characterization

Drop weight technique was used to characterize breakage dis-tribution function of the mill feed material so as to reflect breakagecharacteristics to the model parameters of the mill. Breakage testwas conducted on single particles in the size fraction of9.5 + 8 mm at an energy level of 1 kW h/ton. A modified manualversion of a JK Tech drop weight test device (Napier Munn et al.;Brown and Grimes, 2005) was used in the characterization tests.Specifications of the drop weight tester which was used was givenby Gen (2002) and Gen et al. (2004). It was proposed to use acombined breakage function that was determined by combiningthe single particle impact breakage functions of individual compo-nents of the mill feed using the weight percentages of the mill feedcomponents (Gen and Benzer, 2008) in modelling of cementgrinding mills. The combined breakage function determined onthe basis of the mentioned assumption and is shown in Fig. 9 astotal feed combined. Mill feed is composed of 60% clinker, 24%trass, 11% limestone and 5% gypsum by weight and used to deter-mine the combined breakage function. Combined breakage distri-bution was found to be shifted towards the breakage function ofthe dominant component of the mill feed which was clinker.

Investigated raw meal mill feed constitutes 74% limestone, 24%clay and 2% iron ore by weight. According to the recorded findings(Gen and Benzer, 2008), the approach was to use single particlebreakage distribution function of limestone which is the majorcomponent of the raw meal mill feed to estimate the averagebreakage function and presented in Fig. 9.Table 7Ball sample amounts along the mill length.

Grinding compartment length (m) Total sample weight (kg)

1 26.272 33.143 28.804 15.43Standard Bond work index value of the mill feed material wasalso experimentally determined as 11.03 kW h/ton according toTS 7700 standard (TS 7700, 1989) using a 90 lm test sieve.3.5. Ball mill model

Air-swept ball mill was modelled using the perfect mixing mod-elling approach (Whiten, 1974) which defines the comminutionprocess in terms of three parameters; breakage rate, discharge rateand breakage function Eq. (1). On the other hand, discharge rate(di) of particles were defined to be a function of mill product (pi)and mill hold-up (si) as given by Eq. (2) (Napier Munn et al.)

f i piri=di Xij1

aijpiri=di pi 0 1di pi=si 2

In these equations, fi and pi are the mass flowrates (t/h) of sizefraction i in mill feed and product respectively, aij is the breakagefunction (in the form of single column step triangular matrix), riis the specific breakage rate of size fraction i (tonnes broken perhour per tonne in the mill which is h1), di is the specific dischargerate of size fraction (i) (tonnes discharged per hour per tonne in themill which is h1), and si is the mass of size fraction (i) inside the

Fig. 11. Specific breakage rates (ri) in the air-swept raw meal ball mill. Replottedafter (Gen et al., 2008).

. Gen /Minerals Engineering 74 (2015) 4150 47mill as tons. Perfect mixing model was used by Benzer (2004) inmodelling of an air-swept raw meal grinding ball mill by consider-ing the single compartment mill as three perfectly mixed tankswhereas air-sweeping through the mill was modelled by a classi-fier at the mill discharge. In the model, tank-1 corresponded tothe mill length where lifting liners were applied whereas tank-2and tank-3 lengths corresponded to the mill length where classify-ing liners were applied. In the related study, mill performance wasevaluated through particle size versus r/d combined breakage rateparameter which normalized the discharge rate effect.

In order to correct the variations in residence time, di is scaledin terms of the mill volume and volumetric feed rate (Q) to theterm di using Eq. (3), where D and L are the diameter and thelength of the mill respectively. Then, r/d model parameter is calcu-lated. Normalized discharge rate (di ) is a function of particle sizeEq. (3) (Napier Munn et al.)

di di

4Q=D2L3

Normalized discharge rate function variation established usingthe estimated mill hold-up (si) was given for the investigated air-swept raw meal mill in Fig. 10. Experimentally determined values(measured) are denoted by the scatter plot and compared with thetypical trend observed in semi-autogenous grinding mills (SAG)(Napier Munn et al.; Leung, 1987) which is denoted by the dottedlines in Fig. 10. This function was calculated by eliminating theclassification effect of the discharge grate. The discharge rate func-tion (di) was considered to be the product of two mechanisms;transport and classification by the discharge grate as explainedfor SAG mills by Leung (1987).

There is a critical particle size in the mill which is denoted by xcand can be determined using normalized discharge rate (di ) func-tion as shown in Fig. 10. Particles finer than this size (xc) behavelike a fluid medium in the mill and discharge at a constant ratethrough the mill. The rate of discharge for particles coarser thanthis size was found to decrease systematically in wet grindingconditions (Napier Munn et al.; Morrell and Man, 1997). Particlescoarser than the grate size (xg) remain in the mill for further sizereduction where the discharge rate equals to zero. In the investi-gated air-swept mill, the fluid medium corresponded to air andthe critical particle size (xc) was expected to be highly dependedon the airflow rate through the mill (Fig. 10).

In this study, the modelling approach was to consider the millas a perfectly mixed single tank as the whole length of the millwas lined with classifying liners. Specific discharge rate functions(di) were calculated from Eq. (2) using the estimated millhold-up. Specific breakage rate (ri) function was estimated usingthe calculated discharge rate functions from Eq. (1). Mill hold-up(tons in each size fraction) in grinding compartment wasFig. 10. Measured normalized discharge rate function (di ) in a full scale fully air-swept raw meal mill (xc = 50 lm). Replotted after (Gen et al., 2008).calculated using the size distribution of average mill content andmeasured mill filling data at the crash-stop condition. The specificbreakage rate calculation procedure was formulated on Excel

spreadsheets.Specific breakage rate function is presented in Fig. 11. Agree-

ment between experimental and back-calculated mill product sizedistributions are given in Fig. 12. The experimental data was foundto be fitted to the model satisfactorily. Specific breakage rates wereassumed to not change along the mill in the modelling approach.

The r/d combined breakage rate parameters of the perfect mix-ing model were calculated as ln(r/d) in the model fit module of theJKSimMet simulator considering the mill as a perfectly mixed sin-gle tank. The fitted values were of the best values that defined themill discharge size distribution. The r/d breakage rate parametersfitted to the perfect mixing model were tabulated in Table 8 andused in the simulation step which characterized the specific break-age rates in the mill. It should be mentioned that, spline functionknot values, which could be defined usually by maximum of fourdata points, were selected from the whole set of specific breakagerate values calculated for each particle size given in Fig. 11.

3.6. Static separator model

Grinding efficiency in ball mills depends on the classifying per-formance of air separators as explained in the study of (Klumparand Slavsky, 1989) and (Kolacz, 1999). Their findings indicatedthat, energy consumption in ball milling can be reduced if the clas-sification efficiency is sufficiently high. The classification behaviorof air separators are described using the efficiency curve concept inthe literature (Austin et al., 1975; Zhang et al., 1988; Zhang, 1992;Benzer, 2000; Luckie and Austin, 1975; Schneider et al., 1983;Kuhlmann, 1984; Dunn, 1985; Plank, 1985; Kellett and Rock,1986; Benzer et al., 2001; Hashim, 2003; Gnl, 2006; Altun,2007). The mathematical equation of the efficiency curve modelis given in Eq. (4) (Napier Munn et al.).

Eoa C1 bbxexpa 1expabx expa 2

4

where,Eoa: fraction of feed reporting to overflow.C: fraction undergoing real classification (1-bypass fraction).a: reduced efficiency curve sharpness parameter.b: reduced efficiency curve fish hook parameter.b: parameter to preserve the definition d50c, i.e. d = d50c whenE = (1/2)C where E denotes the fraction of feed.x: ratio of particle size d to corrected size d50c.

Fig. 12. Agreement between experimental and calculated (model fitted) millproduct size distributions.

Table 8ln(r/d) combined model parameters of the ball mill.

Particle size (mm) ln(r/d)

1.18 4.000.425 1.780.15 0.830.045 0.23

Fig. 13. Efficiency curve (tromp) for static separator (d50 = 0.099 mm; by-pass = 11.85%; fish-hook = 2.39%).

Table 9Model fitted efficiency curve parameters used in the simulation of circuit.

Model parameter Value

d50c 0.1069C (1-by-pass) 85.15a 3.74b 0.3633b 1.16

48 . Gen /Minerals Engineering 74 (2015) 4150d50c: size of a particle in feed which has equal probability ofgoing to underflow or overflow (cut size)

The fraction of feed reporting to underflow (EUA) was defined as1Eoa (Napier Munn et al.). The separator performance can bemodelled in terms of d50c, C, a and b. It was stated that, b controlsthe initial rise in the efficiency curve at fine sizes, while a deter-mines the slope at larger values of d which is around d50c. b is cal-culated iteratively during the fitting of Eq. (4) Whiten, 1966. Effectsof operational parameters on efficiency curve model parameterswere given for air separators used in the cement industry byGnl (2006), Altun (2007) and Benzer et al. (2001). The efficiencycurve (tromp curve) for the static separator established on thebasis of the mass balanced size distributions is presented inFig. 13. The characteristic efficiency curve parameters which ared50, by-pass and fish-hook are also given in Fig. 13.

Fish-hook parameter characterizes the difference between themaximum percentage of fine material amount that appears incoarse stream (underflow of the separator) and the by-pass per-centage. Model fitted efficiency curve parameters used in the sim-ulation of the circuit are given in Table 9. The separatorperformance is not at maximum as 11.85% of feed reports to sepa-rator coarse product. However, this value is reasonable and classi-fication performance of the static separator is sufficiently high.4. Simulation

Simulation model of the circuit was designed in simulationmodule of the JKSimMet simulator by defining the perfect mixingmodel parameters of the air-swept ball mill and efficiency curvemodel parameters of the static separator given in Tables 8 and 9respectively. The ball mill was simulated as a single compartmentmill by eliminating the mill length of 2.935 m which was used indrying stage, such that the full length (L = 9.87 m) of the mill wasused in grinding. Thus, drying of the raw meal outside the millby an appropriate dryer was assumed. Static separator perfor-mance was sufficiently high and assumed to not change at the sim-ulated condition. Cyclones are used to separate static fines fromgas and store static separator fine product (cement). There is notany classification. Thus, cyclones were excluded in the simulationmodel whereas electrofilter return was identified as a stream.The circuit response to the proposed operational condition interms of tonnage flow rates and fineness (0.045 mm passing per-centage) is presented in Table 10. Mass balanced particle size dis-tributions in comparison to those obtained after simulation(simulated) at 23% capacity increase case in the cement through-put are given in Fig. 14. Simulation parameters were kept constantduring the optimization study.

As a consequence of the proposed modification in the mill andthe expected capacity increase, a series of operational modifica-tions will be required such that, regulation of static separator oper-ational parameters. For instance, particle size distribution of thestatic separator feed (mill discharge) is estimated to become fineras indicated by the simulated particle size distributions which willrequire the optimization of the static separator. Parameters thatcan be adjusted in the classification process to attain the targetfineness were recorded in the literature by Kohlhaas (1983) as:

varying of the air flow rate; increase in air flow rate willdecrease the cut size (d50),

adjusting of the deflector over the bottom of the inlet ductthrough which the powder carrying air enters the separator;position of the deflector can be adjusted which will effect thecut size (d50),

adjusting of the top outlet duct; where the cut size can be var-ied by vertical adjustment of the air outlet duct at the top of theseparator. For a constant air flow rate, increase in the length ofthe duct will lead to decrease the cut-size (finer product) or viceversa.

Air flow rate in the duct of the mill should be increased beforethe adjustment of the static separator parameters (i.e., angle set-ting adjustable vanes, deflector) by controlling the by-passamount. The cyclone performance will change depending on thecyclone geometry such that, as the cyclone diameter decreasesand the length of the conical section increases, centrifugal force

Table 10Comparison of crash-stop and simulated cases.

Crash-stop condition (Calc) Simulated condition (Sim)

Stream flows t/h 0.045 mm passing % t/h 0.045 mm passing %

Total fresh feed 85.13 1.91 105.00 1.91Mill discharge 149.26 52.32 165.27 58.21Static separator reject 64.14 16.29 60.27 21.43Electrofilter return 4.97 100.00 4.97 100.00Final cement 90.10 80.71 109.97 82.41

Fig. 14. Agreement between mass balanced and simulated particle size distribu-tions of streams.

. Gen /Minerals Engineering 74 (2015) 4150 49effect on the particle flow pattern will increase and will lead toeffective separation of powder carrying air as explained byKohlhaas (1983). Separation efficiency of the cyclones willdecrease at very low or high grain concentrations. Based on thesimulation results, cyclone and electrofilter capacities are expectedto handle 23% capacity increase in addition to the increase in thedust concentration in the product cyclone overflow. However, thecyclone will be operated at full capacity.

5. Conclusions

Conventional two-compartment fully air-swept KHD HumboldtWedag raw meal ball mill operating in closed circuit with a staticseparator was modelled and simulated to evaluate the probablecapacity increase in the circuit in case the pre-drying compartmentwas used in the grinding stage. The mill was modelled as a per-fectly mixed single tank as the material discharge was providedonly by air-sweeping. Performance of the separator was assumedto not change in the simulation stage.

Simulation results indicated that, 23% capacity increase in thecement throughput could be achieved at the steady state conditionby operating the pre-drying compartment at the same ball chargelevel and ball size distribution, without any change in the productcyclone capacity, and by assuming that the process of pre-drying isperformed in the ball mill upstream. However, air flowrate throughthe mill should be critically regulated as the velocity of the air con-trols the particle size distribution of the mill product in addition tothe operational parameters of the static separator at the new oper-ational condition for a stable and optimum production rate. Grind-ing heat generated could increase which may lead toagglomeration of particles unless reduced. The new design mayrequire larger dust collectors, larger ventilation fans which willbring additional cost.Acknowledgements

Authors appreciation goes to SET Italcementi Group BalkesirPlant for providing the access to the plant and their valuable sup-port during the sampling survey. Prof. A. Hakan Benzer for his valu-able discussions and contributions, Assistant Prof. Okay Altun andAssistant Prof. Hakan Dndar from Hacettepe University are alsogratefully acknowledged.References

Altun, O., 2007. Comparison of different efficiency curve approaches in modelling ofair classifiers. MSc Thesis, Mining Engineering Department, HacettepeUniversity, Ankara, Turkey.

Apling, A.C., Ergin, H., 1994. Validation of a grinding model for a full-scale air sweptcement mill. Demirel-Ersayn (Ed.), Progress in Mineral Processing Technology,1994, pp. 527538.

Austin, L.G., Luckie, P.T., Wightman, D., 1975. Steady state simulation of a cementmilling circuit. Int. J. Miner. Process. 2, 27150.

Austin, L.G., Klimpel, R.R., Luckie, P.T., 1984. Process Engineering of Size Reduction:Ball Milling. AIME Publication, NY, pp. 205290, 51119.

Benzer, A.H., 2000. Mathematical Modelling of Clinker Grinding Process. PhD Thesis,Hacettepe University, Ankara, Turkey.

Benzer, A.H., 2004. Modelling and simulation of a fully air swept ball mill in a rawmaterial grinding circuit. Powder Technol. 150, 145154.

Benzer, A.H., Ergn, S.L., ner, M., Lynch, A.J., 2001. Simulation of open circuitclinker grinding. Miner. Eng. 14 (7), 701710.

Benzer, H., Ergun, L., Lynch, A.J., Oner, M., Gunlu, A., Celik, I.B., Aydogan, N., 2001.Modelling cement grinding circuits. Miner. Eng. 14 (11), 14691482.

Brown, D., Grimes, A., 2005. JK tech drop weight testing device. Operating Manual,JKMRC, The University of Queensland, Brisbane, Australia.

Duda, W.H., 1985. Cement-Data-Book, International Process Engineering in theCement Industry, vol. 1,, French & European Publication, pp. 80656.

Dunn, M.R., 1985. A method for analysing the performance of a mechanical airseparator. World Cem., 327332

Ergin, H., 1993. The Simulation of an Air Swept Cement Grinding Mill. PhD Thesis,University of Leeds.

Gen, ., 2002. An Investigation of the Breakage Distribution Functions of Clinkerand Additive Materials. MSc Thesis, Hacettepe University, Mining EngineeringDepartment, Turkey (in Turkish).

Gen, ., 2008. An Investigation on the Effect of Design and Operational Parameterson Grinding Performance of Multi-compartment Ball Mills used in the CementIndustry. PhD Thesis, Hacettepe University, Mining Engineering Department,Turkey (in English).

Gen, ., Benzer, A.H., 2008. Analysis of single particle breakage characteristics ofcement clinker and cement additives by drop-weight technique. J. ChamberMin. Eng. Turkey 47 (1), 1326.

Gen, ., Benzer, A.H., 2009. Single particle impact breakage characteristics ofclinkers related to mineral composition and grindability. Miner. Eng. 22 (13),11601165.

Gen, ., Benzer, A.H., Ergn, S.L., 2004. Single particle breakage characterization ofmaterials by drop weight testing. Physicochem. Probl. Miner. Process. Pol. 38,241255.

Gen, ., Benzer, A.H., Ergn, S.L., 2008. Effect of ball size classification on specificbreakage rates of particles in full-scale cement grinding tube mills. In: 24thInternational Mineral Processing Congress Proceedings, Beijing, China, pp. 529539.

Gnl, A., 2006. Mathematical modelling of air separators used in cement grindingcircuits. PhD Thesis, Mining Engineering Department, Hacettepe University,Ankara, Turkey.

Hashim, S., 2003. Mathematical modelling the two-compartment mill andclassification. PhD Thesis, Julius Kruttschnitt Mineral Research Centre, TheUniversity of Queensland, Australia.

Kellett, D., Rock, H.G., 1986. Operating experience with the O-Sepa air separator.Zement-Kalk-Gips, (Translation of No. 6/86), 234235.

Klumpar, I.V., Slavsky, S.T., 1989. Designing and testing air classifiers. Int. J. Eng.Fluid Mech. 2 (2), 119140.

Kohlhaas, B., 1983. Cement Engineers Handbook, fourth English ed.

http://refhub.elsevier.com/S0892-6875(15)00007-2/h0030http://refhub.elsevier.com/S0892-6875(15)00007-2/h0030http://refhub.elsevier.com/S0892-6875(15)00007-2/h0035http://refhub.elsevier.com/S0892-6875(15)00007-2/h0035http://refhub.elsevier.com/S0892-6875(15)00007-2/h0075http://refhub.elsevier.com/S0892-6875(15)00007-2/h0075http://refhub.elsevier.com/S0892-6875(15)00007-2/h0185http://refhub.elsevier.com/S0892-6875(15)00007-2/h0185http://refhub.elsevier.com/S0892-6875(15)00007-2/h0185http://refhub.elsevier.com/S0892-6875(15)00007-2/h0205http://refhub.elsevier.com/S0892-6875(15)00007-2/h0205http://refhub.elsevier.com/S0892-6875(15)00007-2/h0025http://refhub.elsevier.com/S0892-6875(15)00007-2/h0025http://refhub.elsevier.com/S0892-6875(15)00007-2/h0170http://refhub.elsevier.com/S0892-6875(15)00007-2/h0170http://refhub.elsevier.com/S0892-6875(15)00007-2/h0130http://refhub.elsevier.com/S0892-6875(15)00007-2/h0130http://refhub.elsevier.com/S0892-6875(15)00007-2/h0130http://refhub.elsevier.com/S0892-6875(15)00007-2/h0100http://refhub.elsevier.com/S0892-6875(15)00007-2/h0100http://refhub.elsevier.com/S0892-6875(15)00007-2/h0100http://refhub.elsevier.com/S0892-6875(15)00007-2/h0125http://refhub.elsevier.com/S0892-6875(15)00007-2/h0125http://refhub.elsevier.com/S0892-6875(15)00007-2/h0125http://refhub.elsevier.com/S0892-6875(15)00007-2/h0125http://refhub.elsevier.com/S0892-6875(15)00007-2/h0180http://refhub.elsevier.com/S0892-6875(15)00007-2/h0180http://refhub.elsevier.com/S0892-6875(15)00007-2/h0150http://refhub.elsevier.com/S0892-6875(15)00007-2/h0150

50 . Gen /Minerals Engineering 74 (2015) 4150Kolacz, J., 1999. Control of the mill charge behaviour in dry tumbling mills. Miner.Eng. 12 (1), 5164.

Kuhlmann, K., 1984. Significance of classification in cement grinding results of abalance analysis of grinding circuit. Zement-Kalk-Gips (Translation of No. 9/84),257260.

Leung, K., 1987. An energy-based ore specific model for autogenous and semi-autogenous grinding. PhD Thesis, JKMRC, The University of Queensland,Australia.

Luckie, P.T., Austin, L.G., 1975. Mathematical analysis of mechanical air separatorselectivity curves. Trans. IMM 84, 253255.

Morrell, S., Man, Y.T., 1997. Using modelling and simulation for the design of fullscale ball mill circuits. Miner. Eng. 10 (12), 13111327.

Napier Munn, T.J., Morrell, S., Morrison, R.D., Kojovic, T., 2005. Mineralcomminution circuits their operation and optimization. JKMRC MonographSeries in Mining and Mineral Processing, No. 2, The University of Queensland,Brisbane.

Plank, F.W., 1985. Influence of the separating effect of the classifier on the capacityof the closed-circuit grinding plant. Zement-Kalk-Gips, (Translation of No, 1/85), 6265.

Polysius, 2002. Tube mills for dry grinding. Krupp Polysius ManufacturersCatalogue.Schneider, L.T., Eickholt, H., Blasczyk, G., 1983. Influence of separators on grindingsystems operating results. Zement-Kalk-Gips (Translation No. 12/83), 2631.

Smidth, F.L., 2002. Tirax Raw Mill. F.L. Smidth Manufacturers Catalogue.TS 7700, 1989. Determination Method of Grinding Work Index, Ankara.Viswanathan, K., 1986. Computer based models for grinding and industrial case

studies. Aufbereitungs Technik 27, 560572.Viswanathan, K., Narang, K.C., 1988. Computer simulation and optimization of ball

mills/circuits. World Cem. 19, 143148.Viswanathan, K., Reddy, C.S., 1992. Mathematical modelling of the effect of air

ventilation on solids recycle rate and ball mill performance. World Cem., 2026Whiten, W.J., 1966. Winter School on Mineral Processing. Dept. of Min. and Met.

Eng, The University of Queensland, Australia.Whiten, W.J., 1972. Simulation and Model Building for Mineral Processing. PhD

Thesis, The University of Queensland, Australia.Whiten, W.J., 1974. A matrix theory of comminution machines. Chem. Eng. Sci. 29,

588599.Zhang, Y.M., 1992. Simulation of Comminution and Classification in Cement

Manufacture. PhD Thesis, JKMRC, The University of Queensland, Australia.Zhang, Y.M., Napier-Munn, T.J., Kavetsky, A., 1988. Application of comminution and

classification modelling to grinding of cement clinker. Trans. Inst. Min. Metall.Sect. C 97, 207213.

http://refhub.elsevier.com/S0892-6875(15)00007-2/h0085http://refhub.elsevier.com/S0892-6875(15)00007-2/h0085http://refhub.elsevier.com/S0892-6875(15)00007-2/h0165http://refhub.elsevier.com/S0892-6875(15)00007-2/h0165http://refhub.elsevier.com/S0892-6875(15)00007-2/h0165http://refhub.elsevier.com/S0892-6875(15)00007-2/h0155http://refhub.elsevier.com/S0892-6875(15)00007-2/h0155http://refhub.elsevier.com/S0892-6875(15)00007-2/h0145http://refhub.elsevier.com/S0892-6875(15)00007-2/h0145http://refhub.elsevier.com/S0892-6875(15)00007-2/h0175http://refhub.elsevier.com/S0892-6875(15)00007-2/h0175http://refhub.elsevier.com/S0892-6875(15)00007-2/h0175http://refhub.elsevier.com/S0892-6875(15)00007-2/h0160http://refhub.elsevier.com/S0892-6875(15)00007-2/h0160http://refhub.elsevier.com/S0892-6875(15)00007-2/h0040http://refhub.elsevier.com/S0892-6875(15)00007-2/h0040http://refhub.elsevier.com/S0892-6875(15)00007-2/h0045http://refhub.elsevier.com/S0892-6875(15)00007-2/h0045http://refhub.elsevier.com/S0892-6875(15)00007-2/h0050http://refhub.elsevier.com/S0892-6875(15)00007-2/h0050http://refhub.elsevier.com/S0892-6875(15)00007-2/h0010http://refhub.elsevier.com/S0892-6875(15)00007-2/h0010http://refhub.elsevier.com/S0892-6875(15)00007-2/h0005http://refhub.elsevier.com/S0892-6875(15)00007-2/h0005http://refhub.elsevier.com/S0892-6875(15)00007-2/h0055http://refhub.elsevier.com/S0892-6875(15)00007-2/h0055http://refhub.elsevier.com/S0892-6875(15)00007-2/h0055

Optimization of a fully air-swept dry grinding cement raw meal ball mill closed circuit capacity with the aid of simulation1 Introduction2 Methods2.1 Sampling survey2.2 Experimental

3 Results and discussions3.1 Mass balancing3.2 Mill inside sampling and granulometry3.3 Ball size classification3.4 Material characterization3.5 Ball mill model3.6 Static separator model

4 Simulation5 ConclusionsAcknowledgementsReferences