Minerals Engineering 17 (2004) 12691279 This article is also available online at: www.elsevier.

com/locate/mineng

Methodologies for the evaluation of grinding media consumption rates at full plant scale
lveda Jaime E. Sepu
Moly-Cop Grinding Systems, Av. Pedro de Valdivia, Santiago, Chile Received 3 May 2004; accepted 8 August 2004

Abstract Due recognition of the signicant impact of the cost of grinding media on the overall economics of comminution facilities worldwide has created the need for reliable, practical methodologies to compareover extended control periodsdierent operating conditions, arising from eventual changes in ore type, ball supplier or simply, a trial of new products from the same supplier. The current publication describes the main aspects to be considered in the planning and execution of a full scale evaluation campaign, the actual data to be recorded and the required calculation routines, including the theoretical framework justifying their applicability. The Linear Wear Model, herein described, provides a theoretical framework for the best estimations of comparative grinding media wear performance (in the absence of ball breakage) in any given application, on the basis of the Specic Wear Rate Constant, kE d (lm/(kW h/ton)), derived from the Specic Consumption Rate (gr/kW h), corrected by actual make-up ball size (mm). When signicant ball breakage is to be expectedas in semiautogenous grinding (SAG) applicationsan expanded, conceptual model, based on pilot Drop Ball Testing (DBT) results is gaining acceptance as a way to incorporate breakage as a possible media consumption mechanism. Ideally, evaluations should be conducted in parallel grinding sections, in order to have the option of establishing multiple sequential, concurrent or cross-reference comparisons. The evaluation period should cover at least 6 months after the complete purge of the string of balls being substituted, which may well take from 3 to 6 months. As a result, the evaluation period required for reliable conclusions should typically exceed 912 months of fairly undisrupted, normal operation. The methodologies here proposed make intensive use of Moly-Cop Tools, a software package developed by the international Moly-Cop Grinding Systems organization (formerly, ARMCO Worldwide Grinding Systems) with the specic purpose of helping process engineers characterize and evaluate the operating eciency of any given grinding circuit, following standardized methodologies and widely accepted evaluation criteria. 2004 Elsevier Ltd. All rights reserved.
Keywords: Comminution; Grinding mills

1. Introduction Primary and secondary grinding circuit operators worldwide are fully aware of the signicant impact of grinding media consumption on the overall cost structure of any particular, full scale grinding facility. The

E-mail address: [email protected]. 0892-6875/$ - see front matter 2004 Elsevier Ltd. All rights reserved. doi:10.1016/j.mineng.2004.08.007

cost associated with this essential consumable depends mainly on two factors: delivered price and durability (quality) of the grinding media. When comparing dierent operating conditions, arising from eventual changes in ore type or ball supplier (or simply, a trial of new products from the same supplier), there is a well accepted evaluation criterion that accounts for both media price and quality, referred to as the Eective Grinding Cost or the Cost-Eectiveness of the application. Any

1270

lveda / Minerals Engineering 17 (2004) 12691279 J.E. Sepu

given operating condition is considered to be cost-eective when its unit grinding media costnormally expressed in $/ton groundis to some extent reduced, with respect to a nominal reference condition: grinding media cost $=ton ground ball price $=ton balls ball consumption ton balls=ton ground 1

dd =dt 2k m =qb k d

where: d size (diameter) of the grinding body, after t hours in the mill charge (mm) qb density of the grinding body (gr/cm3 or ton/m3) kd linear wear rate constant (mm/h) The above rst-order dierential equation may be easily integrated for the particular and most frequent practical case in which kd remains constant with timethat is, kd is not a function of the instantaneous ball diameter (Linear Kinetics)and the mill is continuously recharged with media of a single size dR (monorecharge). In such case: d d R kdt 4

Under this criterion, an alternative, higher-price grinding media product could be cost-eective if its associated consumption rate is suciently lower than the reference media, to yield an also lower grinding cost, as dictated by Eq. (1). For the proper application of the above criterion, it is then a basic requirement to maintain continuously updated and representative indicators of the performance of any particular grinding media type being utilized. In each particular case, media price should always be a known, well dened variable; however, it is not so evident how media performance (quality) dierences amongst alternative product types or operating conditions could be assessed with reasonable accuracy and precision. In such context, the current publication proposes a functional theoretical frameworkbased on the Linear Wear Model, expanded to incorporate a fairly conceptual, simple Breakage Modelto derive the most representative indicators of grinding media quality in any given grinding application.

indicating that the rate at which the grinding body loses diameter is constant with time. In other words, if a ball loses 1.0 mm in diameter during the rst 100 h in the mill charge, it will also lose 1.0 mm in the second 100 h in the charge, and in the tenth 100 h, and so on, until the ball is totally consumed or rejected from the mill. In those much less frequent caseswhen kd is a function of the instantaneous ball diameter (Non-Linear lveda, Kinetics)more complex, but also available (Sepu 2001a,b), models should be applied. Eq. (4) is the basis for the simultaneous, experimental determination of kd constants of many dierent types of balls, all present in any given mill charge at the same time; a methodology referred to as Marked Ball Wear lveda, 2003). Testing (MBWT) (Sepu 2.2. Application of the linear wear model to full scale mills In order to maintain a constant inventory (hold-up) of grinding media in the millnormally measured by the ratio Jb of the apparent volume of balls (i.e., including interstitial spaces in between the balls) to the total eective internal mill volumeoperators must continuously compensate for the steel being consumed by recharging new media, preferentially of a single size dR. In this regard, the typical practice of recharging on a shift-per-shift or day-per-day basis may well be considered continuous, for all modeling purposes. Regardless of the recharge frequencyas a reection of the linear nature of Eq. (4) abovethe size distribution of the media in the mill charge should approximate the so-called uniform size distribution. For example, consider the case of an operator charging 1000 new balls of 50 mm diameter, every day. By the following day, those balls will have a diameter of say (50 0.1) = 49.9 mm and he will be adding 1000 new balls of 50 mm. By the next following day, the rst charged balls will have a diameter of (50 2 0.1) = 49.8 mm and the balls charged just the day before will have a size of

2. Referential theoretical framework 2.1. Grinding media wear kinetics The most widely accepted approach to characterize the slow, sustained consumption (wear) kinetics of grinding bodies in rotary tumbling mills is known as the Linear Wear Theory (Prentice, 1943; Norquist and Moeller, 1950); according to whichat every instant t after the grinding body was thrown into the mill chargeits rate of weight loss will be directly proportional to its surface area exposed to gradual abrasion and/or corrosion wear mechanisms: Xt dm=dt k m Ab where: Xt m Ab km 2

media consumption rate (kg/h) ball weight (kg), after t hours in the mill surface area of the ball exposed to wear (m2) mass wear rate constant (kg/h/m2)

Equivalently, taking into account the geometry of the grinding body (sphere or cylinder), Eq. (2) converts to

 lveda / Minerals Engineering 17 (2004) 12691279 J.E. Sepu

1271

(50 0.1) = 49.9 mm. So, after 3 days, not accounting for all the older media, the mill will contain 1000 new 50 mm balls, 1000 almost new 49.9 mm balls and 1000 not so new 49.8 mm balls, assuming kd remained constant at 0.1 mm/day. If this practice is maintained for a suciently large number of days, the operator will nd that, in the same period that he recharges another 1000 new balls, he will be losing from the mill charge 1000 old ball cores or nuclei that reached a small enough critical rejection (scrap) size. From then on, the mill will contain a string of equal number (1000 balls) of every possible size; that is, a uniform distribution of sizes that can be mathematically described by the simple expression: f0 d 1=d R d S
S

[dS, dR]demonstrates that the overall grinding media consumption rate Xt (kg steel/operating hours), corresponding to the ensemble of balls in the mill charge, is consequently proportional to the total area exposed A (m2) by the string: Xt k m A qb k d A=2
R

10

For a monorecharge policy, with balls of size d , the total area of the charge may be obtained from the expression: Z dR A pd 2 Nf 0 d dd 11
dS

for d S < d < d R

which upon substitution of Eqs. (5) and (7) plus integration yields a simple expression for the so called Specic Surface Area of the string: a A=V ap 80001 fv d R 3 d S 3 =d R 4 d S 4 12

where d (mm) represents the scrap or rejection size, characteristic of the design and operating conditions of each particular application. In Eq. (5), the density function f0(d) is such that f0(d)d(d) = d(d)/(dR dS) represents the number fraction of balls in the mill whose size falls in the innitesimal range [d,d + d(d)]. The mass size distribution F3(d), corresponding to the fraction of the total weight Wb (not number) of balls in the string with size smaller than d, may be determined from the population balance relationship: Z d W b F 3 d qb pd 3 =6Nf 0 d dd 6
dS

where: a specic surface area of the charge, m2/m3, (apparent) Vap apparent mill volume occupied by the charge (including interstitial spaces), m3, calculated as Wb/qb/(1 fv) Wb total weight of balls in the charge (ton) fv volumetric fraction of interstitial voids; typically 3540% According to this equation, the total surface area exposed by the mill charge is inversely proportional to the make-up ball size. On this basis, smaller balls of same intrinsic quality (i.e., same value of kd) will wear o faster just because of the correspondingly larger total surface area exposed. Therefore, when comparing two alternative operating conditions, any observed dierence in Xt does not imply the same corresponding dierence in media quality if the balls are not of exactly the same size. In those special cases when two dierent ball sizes (d R 1 and d R 2 ) are continuously charged to the mill, in a proportion r1:r2 (by weight), the combined area exposed so generated may be calculated with the expression: a A=V ap v1 a1 1 v1 a2 where v1 r1 a2 =1 r1 a1 r1 a2 and the specic areas a1 and a2 are obtained from Eq. R R (12), for d R d R 1 and d d 2 , respectively. Eq. (13) arises from recognizing that each make-up size generates its own independent string and that, in order for them to be consumed in the r1:r2 (by weight) proportion, their total exposed areas (m2) must be in the same proportion inside the mill. 13

The (Wb/N) ratio is derived from the integration of Eq. (6) over the whole range of sizes [dS, dR] imposing that, by denition, F3(dR) = 1.0 to obtain W b =N qb p=24d R d S =d R d S
4 4

Then, upon proper substitution and integration Eq. (6) reduces to F 3 d d 4 d S =d R d S for d S < d < d R 8 In the special case, when dS ! 0, the above equation reduces to F 3 d d =d R
4 4 4 4

for 0 < d < d R

Regarding this simpler case, it is interesting to mention that the well known F.C. Bond (Bond, 1961) empirically determined and proposed a value of 3.8 for the exponent in the above equation; which is fairly close to 4.0, the theoretically derived value. Further, the relatively high value of the exponent in Eq. (9) indicates that most of the weight of balls in the charge is distributed in sizes not much smaller than the original make-up size, dR. Given that the wear rate of each grinding body is proportional to its own exposed surface area, the integration of Eq. (2)over the whole range of sizes

1272

lveda / Minerals Engineering 17 (2004) 12691279 J.E. Sepu

Finally, substitution in Eq. (10) above yields Xt 4000k d qb 1 fv V ap d d S =d R d S or equivalently, Xt 4000k d W b d R d S =d R d S

rrez and Sepu lveda, 1986; Fuerstenau, 1972; Gutie lveda (2001a,b)) model serves such purpose well: Sepu 14 P net gP gross 0:238D3:5 L=DN c qap J 1:065J 2 sin a 18

15

On this basis, kd may be easily back-calculated from actual operating records or estimates of Xt, Wb, dR and dS . Again, as dS ! 0, Eq. (15) reduces to the much simpler form: Xt 4000k d W b =d R 16

which re-emphasizes the fact that smaller balls will produce higher consumption rates (kg/h) for the same ball quality, when characterized by kd.

3. The specic energy as a controlling wear kinetics parameter By direct analogy to mineral particle breakage kinetics, it appears reasonable to postulate that an even more representative and scaleable quality indicator than kd is the Energy Specic Wear Rate Constant [k E d , lm/(kW h/ ton)], dened through the expression: kd kE d P b =W b =1000 17

where: Pgross gross power draw of the mill (kW) = Pnet/g g overall mechanical and electrical transmission eciency (/1) D eective internal diameter of the mill (ft) L eective internal length of the mill (ft) Nc rotational mill speed; expressed as a fraction (/1) of its critical centrifugation speed: Ncrit = 76.6/D0.5 J apparent mill lling (/1) (including balls, rocks (if any), slurry and the interstitial spaces in between the balls and the rocks, with respect to the total eective mill volume) a charge lifting angle (denes the dynamic positioning of the center of gravity of the mill load (the kidney) with respect to the vertical direction, typically with values in the range of 3545 and where qap denotes the apparent density of the charge (ton/m3), which may be evaluated on the basis of the indicated charge components (balls, rocks and lveda, 2001a,b): interstitial slurry) (Sepu qap f1 fv qb J b 1 fv qm J J b qp J p fv J g=J 19 with: fv volume fraction (/1) of interstitial voids in between the balls (typically assumed to be 35 40% of the volume apparently occupied by the balls) apparent ball lling (/1) (including balls and the interstitial voids in between such balls) interstitial slurry lling (/1), corresponding to the fraction of the available interstitial voids (in between the balls and rocks in the charge) actually occupied by the slurry of ner particles mineral particle density (ton/m3) slurry density (ton/m3), directly related to the weight % solids of the slurry (fs) by 1/[(fs/ qm) + (1 fs)]

where the power intensity ratio (Pb/Wb) corresponds to the contribution to mill net power draw Pb (kW) of every ton of balls in the charge (Wb) to the total net power draw Pnet (kW) of the mill. The underlying theoretical claim is that grinding balls will wear faster in a more power intensive environment. In other words, k E d is equivalent to kd, but proportionally corrected by how much power is being absorbed by each ton of balls in the charge. Therefore, it is to be expected that k E d should be more insensitive than kd to variations in mill operating conditions (that may aect Pb and/or Wb) that may, in turn, produce higher or lower media consumption rates (kg/h), not caused by variations in grinding media quality. As a practical evaluation criterion, it should then be accepted that the top quality grinding media, in any given application, will be the one that exhibits the lowest value of the Energy Specic Wear Rate Constant k E d , regardless of the mill operating conditions. Due application of Eq. (17) creates the need for a mathematical representation of the total Net Power Draw of the mill in terms of its main dimensions and basic operating conditions. And also, how each component of the mill charge (balls, rocks (if any) and slurry) contributes to this total net power demand. An expanded version of the simple Hogg and Fuerstenau (Hogg and

Jb Jp

qm qp

Substitution of Eq. (19) into Eq. (18) allows for the decomposition of the total net power draw of the mill, lveda, in terms of the charge components (Sepu 2001a,b). In particular, the contribution by the balls in the charge becomes P b 1 fv qb J b =qap J P net 20

Similarly, the contribution to the net mill power by the rocks in the charge becomes

 lveda / Minerals Engineering 17 (2004) 12691279 J.E. Sepu

1273

P r 1 fv qm J J b =qap J P net

21

and nally, just for completeness, the contribution of the slurry in the charge becomes P s qp J p fv J =qap J P net 22

4. Mathematical characterization of impact breakage as a mechanism for grinding media consumption In operations where noticeable ball breakage is to be expectedlike in high-impact, semiautogenous grinding (SAG) applicationsan expanded, conceptual model, based on pilot Drop Ball Testing (DBT) results has been proposed to incorporate breakage as a potentially significant grinding media consumption mechanism. 4.1. Assessment of impact breakage resistance at pilot scale The Drop Ball Tester (DBT) is a standard, pilot scale testing procedure, originally designed by the US Bureau of Mines (Blickenderfer and Tylczak, 1983, 1985) and later adapted by the international MolyCop Grinding Systems organization to assess the resistance of any given sample or lot of balls to repeated severe ball-to-ball impacts. Briey, the DBT facility consists of a 10 m-high, J-shaped tube of slightly larger internal diameter than the size of the balls being tested. The curved, bottom part of the tube is lled with a constant number of balls (for instance, 24 when testing 5 in. balls). When another ball is dropped through the tube from a height of 10 m above, the top ball retained below in the tube suers the direct impact of the falling ball, which is replicated through the whole line of balls retained in the curve at the bottom of the J-tube, originating the removal (through the lower tip of the Jtube) of the rst ball in the line, which is so replaced by the last ball dropped. The balls removed from the tube are continuously liftedvia a bucket elevator back to the top of the tube to be dropped down once again. The DBT is run until a certain maximum number of balls are broken (say, 5 balls) or a reasonable number of total cycles have been completed (say, 20,000 drops). 4.2. Maximum impact energy in full scale mills Fig. 1 illustrates the most critical, outer trajectory of a ball being lifted to a position dened by the angle /1 in the upper-right quadrant of the section of a mill of diameter D (ft) and then allowed to free-fall down to impact the toe of the mill charge kidney (or worse, to impact any exposed mill lining part) at a position /2, in the left-lower quadrant. The mill is rotating at a speed N, related to its critical centrifugation speed Nc by the expression (see Eq. (18)): N N c 76:6=D0:5 28

Referring back to Eqs. (15) and (17), an additional formula for the Energy Specic Media Consumption Rate, XE (grs of steel/kW h drawn), may now be derived: XE 1000Xt =P b equivalent to XE
R 3 4000k E d d

23 d =d d
S 3 R 4 S 4

24

On this basis, k E d may be easily back-calculated from actual operating records or estimates of XE, dR and dS; recalling that the top quality grinding mediain any given applicationwill be the one that exhibits the lowest value of the Energy Specic Wear Rate Constant kE d , regardless of the mill(s) operating conditions. Again, as dS ! 0, Eq. (24) reduces to
R XE 4000k E d =d

25

which re-emphasizes the fact that smaller balls will produce higher consumption rates (gr/kW h) for the same ball quality, when properly characterized by k E d. 3.1. Purge period When comparing two dierent types of grinding media, an accurate evaluation of their relative performance must necessarily discard all data collected during the so-called purge period of the mill; that is, the lapse of time required for the complete consumption of the last ball charged prior to the beginning of the test with the alternative media or the time required for the complete consumption of the rst new ball charged at the beginning of the test with the alternative media, whichever is longer. Such periods may be readily estimated from Eq. (4), setting d = 0: tmax d R =k d 26

If the test balls are of similar quality, the cumulative grinding media consumption during the whole purge period may be estimated from Eq. (14) above (for dS ! 0): Xt tmax =1000 4qb 1 fv V ap 4W b !!! 27 concluding that the purge period is roughly the time required to consume an amount of steel equivalent to 4 times the tons of balls in the mill load. As a result, a quick calculation of the purge period (months), in any specic application, is simply: [4Wb/(average monthly consumption)]. The resulting value could be as long as 68 months. In practice, it is considered acceptable to purge the mill for the equivalent of only 2Wb as, by that time, there should be no more than 10% of the old string remaining.

Therefore, at the time of release, the ball is moving at a linear speed t (m/s) determined by

1274

lveda / Minerals Engineering 17 (2004) 12691279 J.E. Sepu

Tangential Velocity :
1

= DN
N, rpm

independent of ball size (!). For typical values of Nc = 75%, the above ratio takes on values in the range of 0.61.0 for low and high impact conditions, respectively. 4.3. Projection of DBT results to full scale

Rotational Speed :

N = Nc (76.6/D0.5)

D
Fig. 1. Illustration of the outer, most critical trajectory of a ball in a mill.

The rst necessary step in correlating pilot DBT results to industrial scale performance is to derive estimates of the lifting capacity of the cavities in between lifters; that is, how many m3/h of charge are being lifted up by each cavity; in order to then proceed to assess the number rate of balls being lifted from the kidney and subjected to the most critical impact conditions. With reference to Fig. 2, such volume could be approximated as V L 0:0000983h2 L tan d2 tan d1 L 34
3

t pDN pDN c 76:6=D0:5 0:305=60 0:3894pN c D0:5 29

The free-fall height (m) of the ball is simply determined by Dh 0:305D=2 cos /1 D=2 cos /2 30

where VL (m /lifter) represents the volume of charge being lifted by each lifter bar of height hL (in.), in a mill of eective length L (ft). Therefore, the lifting capacity CL (m3 (app)/h) of the lining system would be C L 0:0059nL Nh2 L tan d2 tan d1 L 35

A high-impact condition would then be represented by angle values around /1 /2 30 and conversely, low-impact conditions by values /1 /2 60. In such context, the maximum energy of a ball of mass m at the time of release from the upper shoulder of the kidney may be estimated as the sum of its kinetic energy plus its potential energy with respect to the point of impact: Emax kinetic energy potential energy mt2 =2 mgDh 0:07581 mp2 N 2 c D 0:305 mg D=2 cos /1 D=2 cos /2 31

where nL represents the number of lifter bars in a cross section of the mill. A proportion Jb/J of CL will correspond to lifted balls: C L;balls m3 app=h J b =J C L In terms of mass of balls: C L;balls ton=h qb 1 fv J b =J C L 37 36

This total energy is assumed to be of a similar magnitude as the maximum energy delivered at impact, not taking into account any rotational kinetic energy or friction losses. The equivalent impact energy imposed in the DBT facility is simply given by EDBT;std mg hDBT;std 32

where hDBT,std (m) represents the free-height of the standard J-tube. Finally, the equivalent DBT height to attain equal impact energy at both scales (pilot and industrial) is obtained from matching equations. (31) and (32) above and recalling that g = 9.806 m/s2: hDBT;eq =D 0:0763N 2 c 0:153cos /1 cos /2 33
Fig. 2. Schematic view of a lifter bar pulling rocks and balls away from the kidney.

 lveda / Minerals Engineering 17 (2004) 12691279 J.E. Sepu

1275

and in terms of number of balls (as per Eq. (8), mostly rather new balls of size close to dR): C L;balls #=h qb 1 fv J b =J C L =qb pd R =10003 =6 38

of grinding media consumption, may be conveniently performed with the aid of the Media Charge_Impact & Wear spreadsheet, included in the Moly-Cop Tools (Se lveda, 2001a,b) software package. pu

This expression allows for the estimation of the total number of balls being lifted (therefore, impacted) per unit of time. But certainly not all of these impacts will be critical enough to cause breakage of the ball. In this regard, they are considered to be critical impacts (i.e., those causing potential breakage) when balls hit balls mostly surrounded by other balls; that is, without the cushioning eect of surrounding rocks. In other words, a ball would have a sizeable probability of breakage when such ball is impacted directly against another ball, surrounded by other balls. And the probability of such breakage is scaled up from the DBT result as DBTstd broken balls=drop # balls in J -tube DBTeq DBTstd hDBT;eq =hDBT;std 39

5. Analysis of full scale results In the analysis of plant scale data, it is not yet normal practice to base comparisons on estimates of wear rate or breakage rate constants like those described in the preceding section. Instead, one or more of the following consumption indicators are used: Consumption by unit of energy consumed, XE (gr/kW h) (see Eq. (24)), Consumption by unit of operating time, Xt (kg/h) (see Eq. (15)), Consumption by unit of ore ground, XM (gr/ton). As indicated above, starting from available plant data and corresponding DBT results, the Media Charge_Im lveda, pact & Wear spreadsheet of Moly-Cop Tools (Sepu 2001a,b) facilitates the simultaneous evaluation of each of these indicators plus the estimate of the more intrinsic wear rate parameters, indicative of the actual grinding media performance in the specic application under consideration. No doubt, the most commonly used indicatorand unfortunately, the least representative of allis the specic consumption rate XM normally expressed in (gr of steel/ton ground), that may be also calculated from the expression: XM gr=ton XE gr=kW h E kW h=ton 41

Furthermore, the probability of a ball directly hitting other ball (and not a rock) is proportional to the (Jb/J) ratio and the probability of a ball being surrounded by other balls is assumed to be also proportional to the same ratio, in such a way that the rate of ball breakage would be estimated by (see Eq. (38)) broken balls=h DBTeq C L;balls #=h J b =J 2 40 which may then be converted into a mass rate (ton/h) consumption of balls. Again, for this latter calculation, it is assumed that all broken balls were of size close to the make-up size, dR. Since [CL,balls (#/h)] is proportional to the (Jb/J) ratio, Eq. (40) shows that this relative lling ratio has a very signicant inuence to the power 3.0. Recent trends in operating practice of many SAG mills worldwide are signaling towards increased ball llings, Jb, at lower total llings, J; that is, increased (Jb/J) ratios. Consequently, higher media consumption rates are to be expected in such cases, for even the same quality of balls. In closing, the underlying claims of the just described theoretical framework are that, in terms of gradual wear resistance, the top quality grinding mediain any given applicationwill be the one that exhibits the lowest value of the Energy Specic Wear Rate Constant k E d, regardless of the mill(s) operating conditions and that, in terms of impact breakage resistance, the top quality media will be the one with the lowest DBTstd parameter. The series of cumbersome calculations involved in the assessment of the combined eect of both mechanisms

where Ethe specic energy consumed (kW h) per each ton of ore grounddepends exclusively on the operating conditions and intrinsic characteristics of the ore, with no relation whatsoever to the quality of the media being used. It is so concluded that any variation in E could be mistakenly interpreted as a variation in media quality, if the XM indicator were to be adopted for comparative evaluations. It is likewise concluded that a much better indicator of media quality is XE, properly corrected by media size dRas suggested by Eq. (24)to arrive at the Energy Specic Wear Rate Constant, k E d [lm/(kW h/ton)].

6. Example of application Consider the hypothetical case of Fair Mining Co. operating two parallel 36 0 / 17 0 SAG lines. Appendix

1276

lveda / Minerals Engineering 17 (2004) 12691279 J.E. Sepu

A contains operational records for both lines in the period July 2002 through June 2003. The standard 5 in. balls are being supplied by Forge+. However, since November 2002, management decided to start an evaluation with the alternative supplier, NKOB. In the interest of evaluating which supplier provides better quality balls, it must be rst recognized that there are multiple ways of establishing such comparison. In fact, discarding 4 months (November 2002 through February 2003) as purge period, at least the following options could be selected: Sequential Evaluation, comparison of historical consumption rates of the same mill (SAG 2), before and after the purge period. Concurrent Evaluation, comparison of consumption rates of a test mill (SAG 2) against a standard mill (SAG 1), both operating in parallel, for exactly the same time period, once the purge period has been completed. In the rst case, the following Sequential Evaluation could be established: SAG 2 Pre Purge Ore throughput ton/h Energy consumption kW (net) kW h/ton Balls consumption gr/ton 1254 12,058 9.62 552 SAG 2 Post Purge 1410 11,691 8.29 501 Variation %

SAG 2 SAG 2 Variation % Pre Post Purge Purge Ore throughput ton/h Energy consumption kW (net) kW h/ton Balls consumption gr/ton kg/h gr/kW h 1254 12,058 9.62 552 692 57.4 1410 11,691 8.29 501 707 60.4 2.95 12.4 (3.0) (13.8) (9.2) 2.2 5.2 5.3

Sp. Wear Constant, k E d 2.80

The additional indicators reveal that, in reality, the alternative media would be, not 9.2% better, but 5.3% worse in performance, as compared to the standard media; proving that XM (gr/ton) is not a reliable media quality indicator, because is distorted by changes in E (kW h/ton) strictly related to process variables and totally independent of media quality. A Concurrent Evaluation, based on the same data base of Appendix A, shows:

12.4 (3.0) (13.8) (9.2) SAG 1 SAG 2 Variation % Post Post Purge Purge Ore throughput ton/h Energy consumption kW (net) kW h/ton Balls consumption gr/ton kg/h gr/kW h 1299 11,791 9.08 479 621 52.7 1410 11,691 8.29 501 707 60.4 2.95 8.5 (0.8) (8.7) 4.6 13.8 14.6 15.4

If the analysis was to be based only on XM (gr/ton), the rst conclusion to be drawn would be that the alternative NKOB balls are 9.2% better than the standard Forge+ balls(?). Even more, the process analyst (or the NKOB vendor) could claim that, thanks to the new balls, throughput was increased by 12.4%! But the analyst should not ignore that there has been a very significant change in E (kW h/ton), before and after the purge period. However, assuming similar DBTstd performance of 5 broken balls every 20,000 drops, for both type of media, the following more detailed comparison may be established, for the same Sequential Evaluation above, using the spreadsheet Media Charge_ lveda, Impact & Wear of Moly-Cop Tools (Sepu 2001a,b):

Sp. Wear Constant, k E d 2.55

that is, much more signicant dierences against the NKOB balls than those indicated by the previous comparison. Which one to believe: the sequential or the concurrent evaluations? There is not a clear cut answer to such question. The merits of the concurrent evaluation is that the comparison is based on the performance of parallel lines, operating for the same periods in time,

 lveda / Minerals Engineering 17 (2004) 12691279 J.E. Sepu

1277

most likely being fed with the same type of ore. However, no two lines in any Concentrator are really identical and in that regard, the sequential evaluation assures that the comparison is not being aected by intrinsic differences in the process equipment. A third option for comparisons, referred to as Cross Reference Evaluation, can be stated in two totally equivalent ways:

have to be explained by an even larger value of k E d associated to the NKOB balls:

SAG 1 Specic Wear Rate Constant, Pre Purge 2.68 Post Purge 2.55 kE d

SAG 2 2.80 3.11 16.4%

Cross Reference A: dierence in the consumption rate of the test mill (SAG 2)before and after the purge periodminus the same dierence for the standard mill (SAG 1), normalized with respect to the standard mill wear constant, before the purge period:
E E k E d SAG2;after k d SAG2;before k d SAG1;after E k E d SAG1;before =k d A;before

42

Or if the NKOB balls were slightly larger, say 130 mm? In such case, going back to the assumption of similar DBTstd performance, the observed dierence in performance would also have to be explained by a larger value of k E d for the NKOB balls:

Cross Reference B: dierence in the consumption rate of both mills (SAG 1 and SAG 2)before and after the purge periodnormalized with respect to the standard mill wear constant, before the purge period:
E E k E d SAG2;after k d SAG1;after k d SAG2;before E k E d SAG1;before =k d A;before

SAG 1 Specic Wear Rate Constant, Pre Purge 2.68 Post Purge 2.55 kE d

SAG 2 2.80 3.02 13.0%

43

For the particular example under consideration, such Cross Reference (A and B are mathematically equivalent) would be

SAG 1 Specic Wear Rate Constant, Pre Purge 2.68 Post Purge 2.55 kE d

SAG 2 2.80 2.95 10.4%

The practice of making balls slightly larger than their nominal size (overweight) is quite common amongst grinding media manufacturers in their continuous eorts to maximize equipment availability and production capacity. It should have no signicant eect on the grinding eciency of the media but, in terms of wear consumption, it could help hide 35% dierences in relative performance, if the consumption indicators are not properly corrected for overweight.

7. Concluding remarks The Linear Wear Model provides a theoretical framework for the most reliable estimations of comparative grinding media wear performance (in the absence of breakage) for any given application, on the basis of the Specic Wear Rate Constant, k E d. Because of natural process uctuations and measurement errors, it is not possible to develop a single comparative performance indicator. In such context, evaluations should be ideally conducted in parallel lines, in order to have the option to establish the preferred Cross Reference comparisons. A fairly simple Ball Breakage Modelof particular interest in SAG applicationsis also available in MolyCop Tools (Version 2), in order to account for this consumption mechanism, whenever suspected signicant.

indicating that whilebecause of a change in ore properties or process conditionsSAG1 was reducing the wear rate, SAG 2 could not do the same because of the dierent grinding media being charged. From a different but equivalent perspective, while SAG 2 was performing slightly worse than SAG 1, prior to the purge period, that dierence was incremented after the purge period because of the lower media quality. In both cases, the most reliable estimate of media quality dierence is 10.4% worse than the standard Forge+ balls in SAG 1 before the purge period. And if the NKOB balls didnt break at all ? In such case, all of the observed dierence in performance would

Appendix A. Example of application Fair Mining Co. Operational Records Unit: SAG1 Mill diam.: 36 ft Mill length: 17 ft % Critical: 76 Ore density: 2.8 ton/m3 Ore Operating Grinding Energy Mill throughput hours apacity consumption power (ton/month) (h/month) (ton/h) (MW h/month) (kW) Jul 2002 Aug Sep Oct Nov Dec Jan 2003 Feb Mar Apr May Jun Average 1,017,541 915,593 908,071 718,227 703,180 852,259 995,836 1,014,800 864,302 935,336 867,843 747,636 878,385 721.0 644.0 715.0 643.0 627.0 695.0 718.0 691.0 639.0 699.0 661.0 631.0 673.7 1411 1422 1270 1117 1121 1226 1387 1469 1353 1338 1313 1185 1304 8533 7639 8576 7506 6960 7712 7872 7814 7606 8231 8071 7103 7802 11,836 11,862 11,994 11,674 11,100 11,096 10,964 11,308 11,903 11,775 12,210 11,256 Balls consumption ton/month gr/ton kg/h gr/kW h Supplier 499.21 375.07 480.04 425.99 358.08 444.01 513.25 464.15 400.83 400.84 436.64 396.00 491 410 529 593 509 521 515 457 464 429 503 530 493 692 582 671 663 571 639 715 672 627 573 661 628 643 58.5 49.1 56.0 56.8 51.5 57.6 65.2 59.4 52.7 48.7 54.1 55.8 55.5 Forge+ Forge+ Forge+ Forge+ Forge+ Forge+ Forge+ Forge+ Forge+ Forge+ Forge+ Forge+ Make-up balls: 5.0 in. / % Balls (nominal): 14% % Charge (nominal): 26% % Solids (nominal): 74

1278 lveda / Minerals Engineering 17 (2004) 12691279 J.E. Sepu

11,581 432.84

Unit: SAG2 Mill diam.: 36 ft Mill length: 17 ft % Critical: 76% Ore density: 2.8 ton/m3 Jul 2002 Aug Sep Oct Nov Dec Jan 2003 Feb Mar 755,228 866,067 845,614 951,123 985,943 701,282 877,346 916,566 915,974 632.0 715.0 691.0 688.0 710.9 549.0 723.2 661.1 678.0 1195 1211 1224 1382 1387 1277 1213 1386 1351 7401 8314 8879 8275 8629 6208 8459 7719 7904 11,711 11,628 12,849 12,027 12,140 11,308 11,697 11,675 11,657 401.29 483.39 494.29 507.00 370.84 470.29 431.54 443.47 412.35 531 558 585 533 376 671 492 484 450 635 676 715 737 522 857 597 671 608 54.2 58.1 55.7 61.3 43.0 75.8 51.0 57.5 52.2

Make-up balls: 5.0 in. / % Balls (nominal): 14% % Charge (nominal): 26% % Solids (nominal): 74 Forge+ Forge+ Forge+ Forge+ NKOB NKOB NKOB NKOB NKOB

 lveda / Minerals Engineering 17 (2004) 12691279 J.E. Sepu

1279

References
Blickenderfer, R., Tylczak, J.H., 1983. A large-scale impact spalling test. WEAR 84, 361373. Blickenderfer, R., Tylczak, J.H., 1985. Laboratory tests of spalling, breaking, and abrasion of wear-resistant alloys used in mining and mineral processing. USBM RI 8968 [Report], pp. 117. Bond, F.C., 1961. Crushing and Grinding Calculations, Part I. Allis Chalmers Co. Publication. rrez, L., Sepu lveda, J.E., 1986. Dimensionamiento y OptimiGutie n de Plantas Concentradoras Mediante Te cnicas de Modelzacio n Matema tica. CIMMChile Publication. acio Hogg, Fuerstenau, 1972. Power relations for tumbling mills. Trans. SME-AIME 252, 418432. Norquist, D.E., Moeller, J.E., 1950. Relative wear rates of various diameter grinding balls in production mills. Trans. AIME 187, 712714. Prentice, T.K., 1943. Ball wear in cylindrical mills. J. Chem. Met. Mining Soc. South Africa (JanuaryFebruary), 99131. lveda, J.E., 2001. A phenomenological model of semiautogenous Sepu grinding processes in a Moly-Cop Tools Environment. In: Proceedings SAG 2001 Conference, vol. 4, Vancouver, BC, Canada, pp. 301315. lveda, J.E., 2001. Moly-Cop Tools, Version 1.0: Software for the Sepu assessment and optimization of grinding circuit performance. Available from: <[email protected]>. lveda, J.E., 2003. Experimental protocol for marked ball wear Sepu tests. Internal Report, Moly-Cop Grinding Systems. Available from: <[email protected]>.

55.0 70.1 65.4 661 799 763 469 597 498 12,033 457.80 11,385 511.82 11,666 534.28 8329 7296 8170 1410 1337 1533 692.2 640.9 700.4 976,000 856,863 1,073,551 Apr May Jun

NKOB NKOB NKOB Average 893,463 673.5 1327 7965 11,828 459.86 515 683 57.7