Minerals Engineering 58 (2014) 100-103 9 Contents lists available at ScienceDirect - - - - M INERALS [NGINEf.RING ·---· -·----- Minerals Engineering - ·--- -- ---- ELSEVIER journal homepage: www.elsevier.com/locate/ mineng ......... The value of automated mineralogy 4] Cross Mark Ying Gu Robert P'. Schouwstra Chris Rule a. b.*, c j/(MRC, The University of Queens land, 40 Isles Road, Indooroopilly, QLD 4068, Australia a Anglo Technical Solutions Research, PO Box 106, Crawn Mines 2025, South Aftica b A11glo Atnerica11 Platinu1n, 55 Marshall Street, Marshalltown, ] oha11nesburg 2001, South Aftica c ARTICLE INFO ABSTRACT Article history: Automated mineralogy methods and tools, such as the Mineral Liberation Analyser ( MLA) and the Received 27 May 2013 QEMSCAN, are now widely used for ore characterization, process design and process optimization. Accepted 17 jar1uary 2014 Several case studies published recently demonstrate that large gains can be obtained throL1gh grinding Available online 13 February 2014 and flotation optimization guided by automated mineralogy data. However, since automated mineralogy can only provide the information pointing to where the process gains can b,e made; it does not directly l(eywords: impact the prodLtction gain. Thus the question is often asked: how to value the contribution of automated Autornated 1ni11eralogy mineralogy to process improvement at a particular plant. This appears to be a difficult question to Value of ir1for1natio11 answer. On close examination however, it is found that this is essentially a qL1estion of the valLme of infor Cor1centrator opti1nisation mation and this is reasonably well docL1mented in various other industries. Hubbard, 2010, in chapter 7 "Measuring the Value of Information'', dealt with exactly this type of problem. The value of information is the redLmced risk of an investment and opportunity loss. The methods HLmbbard developed can be applied to estimate the value of automated mineralogy, as well as metallLmrgical test work, both producing information that reduces the risk of investment.This paper first introduces HL1bbard's theory on the valLte of information and how to measure it. It then applies his methods to estimate the value of automated mineralogy, using Anglo Platinum's fine grinding project as an example. In the end, a general model is developed to allow the simulation of the valLte of automated mineralogy in different mining operations constrained by different parameters. © 2014 Elsevier Ltd. All rights reserved. 1. Introduction It must be emphasized that a utomated mineralogy can only provide information pointing to where the process gains can be Automated mineralogy methods and tools, such as the Mineral m.ade and the extent of the benefit. To realise the gain, the p rocess l iberation Analyser (MIA) and the QEMSCAN, are now widely used ing plant needs to be adjusted and, in most cases, modified signif in the mining indust ry worldwide, with close to two hundred of icantly by expert process engineers. Consequently, the linl< these systems installed in research and company central labs over between automated mineralogy data and final p roduction gain is the last ten years. By measuring samples of ore and processing plant indirect. As such, the question is often asl<ed how to value the con material, these instruments provide statistical distribution of the tribution of automated mineralogy to process improvements at a size distribution a nd associations of minerals of interest, critical particular plant. for ore characterisation, process design and optimization. As such, The answer to the above question is of interest for a t least three automated mineralogy has become an important contributor to pro reas ons. Firstly, a proper valuation of automated mineralogy infor cess mineralogy and geometallurgy. Several case studies of the mation will assist managers to determine the level of investment application of automated mineralogy for process improvements considered necessary to obtain such information. giving t he size were published recently (Rule and Schouwstra, 2011; Rule, 2011; of the mine and operation. Secondly, it will guide the automated MacDonald et al., 2011; Lotter, 2011; Lotter et al., 2010), demon mineralogy profession to produce higher value and more relevant strating significant returns obtainable through grinding and flota information. 'Thirdly, it will encourage the evaluation of process tion optimisation supported by automated mineralogy data. improvement proposals, therefore reducing th.e risl< and maximis ing the return of mining inves tment. * So fa r, there are only qualitative statements in the literature Corresponding autl1or. Tel.: +27 ] 13774602; fax: +27 114961034. regarding the value of automated mineralogy. Quantitative miner E-1nail addresses: [email protected] (Y. Gu), robert.scl:touwstra@a11gloa1nerica11. corn (RP. Scl1ouwstra), [email protected] (C. Rule). alogical data are often used as a diagnostic tool to pin point the l1ttp:/ /dx.doi.org/t 0.10 l 6.Jj.rnine11g.2014.0t .020 0892-6875/© 2014 Elsevier Ltd. All rigl1ts reserved Y. Gu et al. / Minerals Engineering 58 (2014) 100-103 101 proble111 area or u11it i11 a processi11g pla11t. Tl1ere are a 11u111ber of 011ce we l<11ow tl1e value of i11for111atio11, we ca11 use tl1at to case studies sl1owi11g sigi1ifica11t retur11s fro111 opti111isi11g pla11t deter111i11e l1ow 111ucl1 we sl1ould spe11d to obtai11 tl1at i11for111atio11 operatio11 or pla11t desig11 based 011 tl1at i11for111atio11 (Lotter (Hubbard, 2010). 2011 ). Qua11titative 111easure111e11t of tl1e value of auto111ated 111i11- Tl1e above are tl1e esse11tials of Hubbard's tl1eo1y a11d 111etl1ods eralogy l1as 110t bee11 studied to date. 011 tl1e surface, it appears tl1at are releva11t to tl1is paper. It will beco111e clearer wl1e11 we to be a difficult subject a11d vexi11g proble111 to solve. However, it apply tl1e111 to derive a11 effective 111etl1od for 111easuri11g tl1e value is esse11tially a questio11 of tl1e value of i11for111atio11, wl1icl1 is rea of auto111ated 111i11eralogy. so11ably well docu111e11ted i11 various otl1er i11dustries. Hubbard (2010), i11 cl1apter 7 "Measuri11g tl1e Value of I11for111atio11", dealt witl1 exactly tl1is proble111 a11d provided a good fra111eworl< wl1icl1 3. A case study - Anglo Platinum's fine grinding project ca11 be used to esti111ate tl1e value of auto111ated 111i11eralogy. Tl1is paper i11troduces Hubbard's tl1eo1y i11 tl1e co11text of 111i11i11g A11glo Plati11u111, tl1e world's largest pri111a1y producer of plati- i11dustry a11d applies it to esti111ate tl1e value of auto111ated 111i11er 11u111, l1as i11vested substa11tial capital to build tl1e capacity to tracl< alogy, usi11g A11glo Plati11u111's fi11e gi·i11di11g project as a11 exa111ple. tl1e 111i11eralogy of processi11g pla11ts a11d ore sources. After a Afterwards, a ge11eral 111odel is developed to allow tl1e si111ulatio11 detailed study i11to tl1e losses of value 111i11erals (plati11u111 group of tl1e value of auto111ated 111i11eralogy i11 differe11t 111i11i11g opera 111i11erals a11d sulpl1ides ) it beca111e clear tl1at to i111prove recoveries tio11s co11strai11ed by differe11t para111eters. a11 i11crease i11 liberatio11 of tl1e value 111i11erals would be required. After so111e pilot testi11g to de111011strate tl1at fi11er gri11di11g would i11crease tl1e liberatio11 a11d recovery of tl1e value 111i11erals a deci 2. Measuring the value of infonnation - Hubbard's theory sio11 was 111ade to go al1ead witl1 large scale i111ple111e11tatio11 of a fi11e gri11di11g project i11 2009. Tl1e outco111es of tl1e i111ple111e11tatio11 2.1. Measurement and "intangibles" of fi11e gri11di11g at tl1e A111a11delbult operatio11 were publisl1ed i11 detail i11 Rule (2011 ) a11d Rule a11d Scl1ouwstra (2011 ). I11 su111111a1y, Measure111e11t is "A qua11titatively expressed reductio11 of tl1e Al11a11delbult project resulted i11 50%r eductio11 i11 taili11gs grade u11certai11ty based 011 011e or 111ore observatio11" (Hubbard, 2010, witl1 correspo11di11g i11crease i11 recovery of over 5% of plati11u111 p. 23 ). It 11ever co111pletely re111oves u11certai11ty. Based 011 tl1is def group of 111etals (PGM). If a11 i111prove111e11t i11 recovery of 1% i11 i11itio11, 111a11y of tl1e tl1i11gs tl1at were tl1ougl1t to be i111possible PGM could be i111ple111e11ted across all tl1e operatio11s i11 tl1e gi·oup tur11ed out to be quite easy to 111easure. Hubbard (2010, p. 4 ) ob tl1is would tra11slate i11to a gai11 i11 excess of US$75 111illio11 a1111ually served "I11ta11gibles tl1at appear to be co111pletely i11tractable ca11 for Al1glo Plati11u111 (2009 esti111ates based 011 a1111ual productio11, be 111easured. Tl1is 111easure111e11t ca11 be do11e i11 a way tl1at is eco- PGE prices a11d excl1a11ge rate, Rule a11d Scl1ouwstra, 2011 ). Tl1e 110111ically justified." For 111a11y orga11isatio11s, it is a routi11e tasl< to capital cost of i111ple111e11ti11g tl1e fi11e gri11di11g progra111 is US$250 111easure tl1e risl< of ba11l<ruptcy, tl1e value of public l1ealtl1 i11itia 111illio11 a11d a1111ual ru1111i11g cost of tl1e additio11al gi·i11di11g equip- tives a11d tl1e value of IT i11vest111e11ts. Tl1rougl1111a11y case studies, 111e11t is US$15 111illio11. Based 011 tl1ese esti111ates tl1e retur11 of Hubbard fou11d tl1at tl1e perceived i111possibility of 111easure111e11t is capital i11vest111e11t for tl1is i11itiative is witl1i11 six years a11d tl1e ofte11 a11 illusio11 caused by 110t u11dersta11di11g: 11et prese11t value (NPV) of tl1is project is approxi111ately US$158 111illio11 based 011 a1111ual discou11t rate of 10% for 12 years. 1. Tl1e concept of 111easure111e11t - reduce u11certai11ty a11d Tl1is is a good exa111ple to de111011strate l1ow to 111easure tl1e va l1e11ce risl<. lue of auto111ated 111i11eralogy usi11g Hubbard's 111etl1od. Tl1e NPV of 2. Tl1e object of 111easure111e11t - tl1e core of tl1e questio11 a11d US$158 111illio11 is attributed to auto111ated 111i11eralogy a11d 111etal tl1e purpose. l urgical test worl< carried out before a11d duri11g tl1e i111ple111e11ta 3. Tl1e methods of 111easure111e11t - 111odel tl1e u11certai11ty tio11 of tl1e fi11e gri11di11g project. Tl1e cost of tl1e auto111ated statistically. 111i11eralogy a11d 111etallurgical test worl< totalled to less tl1a11 US$2 111illio11/a1111u111, a11 i11sigi1ifica11t a111ou11t co111pared to tl1e value 011ce tl1ose aspects of a proble111 are clearly defi11ed a11d u11der gai11ed. stood, tl1e11 tl1e questio11 beco111es l1ow to best 111easure it. He pro Tl1e value of auto111ated 111i11eralogy is to provide i11for111atio11 posed tl1e followi11g ge11eral approacl1: tl1at reduces tl1e risl< of fi11e gri11di11g project: 1. Model wl1at you l<11ow 110W. 1. By esti111ati11g tl1e pote11tial gai11s of tl1e project1 wl1icl1 effects 2. Co111pute tl1e value of additio11al i11for111atio11. tl1e decisio11 011 l1ow 111ucl1 to i11vest i11 tl1e fi11e gri11di11g project. 3. If eco110111ically justified, co11duct observatio11s tl1at reduce 2. By l1elpi11g tl1e 111etallurgists to deter111i11e l1ow fi11e to gri11d a11d u11certai11ty. to opti111ise tl1e circuit desig11. 4. Update tl1e 111odel a11d opti111ise tl1e decisio11. Let us assu111e tl1at at tl1e ti111e wl1e11 A11glo Plati11u111 co11sidered 2.2. Measure the value of information tl1e fi11e gi·i11di11g project, 111a11age111e11t believed tl1e i11vest111e11t sl1ould l1ave a good cl1a11ce of success (say 80% of tl1e target). Tl1is "I11for111atio11 reduces u11certai11ty about decisio11s tl1at l1ave esti111atio11 is based 011 available auto111ated 111i11eralogy i11for111a eco110111ic co11seque11ces". I11 otl1er words, "I11for111atio11 ca11 reduce tio11 a11d preli111i11ary pilot test worl<. Let us furtl1er assu111e tl1at tl1e u11certai11ty. Reduced u11certai11ty i111proves decisio11s. I111- witl1out tl1e auto111ated 111i11eralogy i11for111atio11, tl1e esti111ated proved decisio11s l1ave observable co11seque11ces witl1 111easurable cl1a11ce of success would be 55%. Tl1is a11d releva11t i11for111atio11 value." It is ge11erally recog11ised tl1at i11for111atio11 l1as value, but are listed i11 Table 1. l1ow to 111easure tl1e value of i11for111atio11 is 110t ge11erally l<11ow11. So, fro111 tl1e i11vest111e11t poi11t of view, auto111ated 111i11eralogy If we ca11 esti111ate tl1e "cl1a11ce of bei11g wro11g a11d tl1e cost of reduced tl1e risl< fro111 45% to 20%. Si11ce we are i11vesti11g US$250 bei11g wro11g" about a decisio11, tl1e11 tl1eir 111ultiplicatio11 is "Ex 111illio11, tl1e expected value of i11for111atio11 (EVI) is US$62.5 111il pected Opportu11ity Loss" (EOL). Tl1e Expected Value of I11for111atio11 lio11 = 250 x (0.45 - 0.20). Fro111 pote11tial gai11 poi11t of view, auto- (EVI) is si111ply tl1e reductio11 of EOL after i11for111atio11 a11d EOL be 111ated 111i11eralogy reduced tl1e risl< of tl1is project bei11g rejected. fore i11for111atio11 available. Tl1e EVI is US$39.7 111illio11=158 x (0.8 - 0.55 ). 102 Y. Gu et al. / Minerals Engineering 58 (2014) 100-103 Table 1 600 • ln1pact of auton1ated nlineralogy (AM) to fine grinding project. Adaption fron1 Hubbard's exhibit 7.1, p. 101. 500 Variable Project works Project fails Chance of success without AM 55% 45% - - - - 400 Chance of success with AM 80% 20% hnpact if project is approved +US$158 1nillion -US$250 1nillion ~ c - - - hnpact if project is rejected -US$158 1nillion US$0 Q) - ::J 300 - - - CT Q...). LL - .... .... - - .... - - 200 1-9- Table 2 The basic operation inforn1ation with calculated returns at a hypothetical plant. - - - . - - .._ - ,_ - .... - - ._ ~ 100 Para1neter Value Note - DT (tonnes) 150,000 Measured OD (days) 360 Measured 0 - ~ ' ' ' ' ' ' AG (%) 1.6 Measured -10 10 30 50 70 90 110 130 150 170 190 210 230 250 MP ($/tonne of contained 1netal) Mean va lue 5000 Esti1natetl Expected Annual Return (Million Dollars) Low value 3000 Esti1nated High value 6000 Esti1nated Fig. 1. Return intervals and frequencies of 5000 si1nulations. t:.R (%) Mean value 2 Esti1nated Low value 0.5 Esti1natecf High value 3.5 Esti1natecf Tl1is is a11otl1er way to lool< at tl1e value of i11for111atio11 provided Return ($M) Mean value US$86 1nillion Calculate'a by auto111ated 111i11eralogy. Low value US$13 1nillion Calculateft Tl1e sa111e ca11 also be applied to esti111ate tl1e value of 111etallur High value US$181 1nillion Calculateft gical test worl< by esti111ati11g 11ow 111ucl1 risl< is reduced by eacl1 DT - daily tonnage of ores processed (tonnes). test project. OD - operation days per year. AG - average grade of ore (%). MP - current 1netal price ($/tonne of contained 1netal in concentrate). 4. Modelling the value of automated mineralogy t:.R - recovery gain= hnproved recovery (IR) - current recovery (CR)(%). Esti1nated values have 90% confidence interval. a Calculated values do not have the sa1ne confidence interval. Its distribution is 111 A11glo Plati11u111's case above, we 111easured tl1e value of auto- b expected to be wider. 111ated 111i11eralogy after tl1e value of 111i11eral processi11g i111prove- 111e11t was realised a11d docu111e11ted. It would be 111ore useful to be able to predict tl1e value of auto111ated 111i11eralogy wl1e11 011ly Table 3 relatively s111all a111ou11ts of i11for111atio11 are available. Tl1is l<i11d Returns and probabilities fron1 5000 sin1ulations. of predictio11 will l1elp 111a11age111e11t to decide 11ow 111ucl1 to i11vest Index Return ($M) Frequency Probability (%) Cu1nulative (%) i11 tl1e auto111ated 111i11eralogy a11d 111i11eral processi11g i111prove- 111e11ts (i.e. pla11t 111odificatio11), so tl1at tl1ey spe11d e11ougl1 to cap 1 10 62 1.24 100.00 2 0 55 1.10 98.76 ture tl1e pote11tial retur11s fro111 i111proved efficie11cy a11d 3 10 128 2.56 97.66 productivity. 4 20 174 3.48 95.10 Tl1e case study above also sl1ows tl1at tl1e value of auto111ated 5 30 242 4.84 91.62 111i11eralogy depe11ds 011 tl1e be11efit of 111i11eral processi11g i111prove- 6 40 314 6.28 86.78 111e11ts. Tl1e i11for111atio11 fro111 auto111ated 111i11eralogy reduces tl1e 7 50 352 7.04 80.50 8 60 479 9.58 73.46 risl< a11d tl1e cl1a11ge i11 EOL. Tl1e size of opportu11ity largely deter- 9 70 485 9.70 63.88 111i11es tl1e co11tributio11 of auto111ated 111i11eralogy. So, wl1e11 we 10 80 430 8.60 54.18 111odel tl1e value of auto111ated 111i11eralogy, we really 11eed to 111odel 11 90 452 9.04 45.58 tl1e retur11 fro111 overall process i111prove111e11ts. 111 tl1is sectio11, we 12 100 405 8.10 36.54 will develop a ge11eral 111odel tl1at would allow us to evaluate tl1e 13 110 354 7.08 28.44 14 120 254 5.08 21.36 retur11 of process i111prove111e11ts for differe11t co111111odities give11 15 130 235 4.70 16.28 a set of pla11t operatio11 co11ditio11s. 16 140 165 3.30 11.58 We 1<11ow tl1at if a cl1a11ge i11 our operatio11 ca11 lead to a11 i11- 17 150 122 2.44 8.28 crease i11 recovery, tl1e11 tl1e value of tl1is cl1a11ge ca11 be calculated 18 160 106 2.12 5.84 19 170 57 1.14 3.72 as follows: 20 180 46 0.92 2.58 Retur11/year =OT x OD x AG x MP x M.wl1ere OT is tl1e daily 21 190 30 0.60 1.66 to1111age of ores processed (to1111es), OD is tl1e operatio11 days per 22 200 21 0.42 1.06 year, AG is tl1e Average grade of ore (%), MP is tl1e curre11t 111etal 23 210 13 0.26 0.64 24 220 5 0.10 038 price ($/to1111e of co11tai11ed 111etal i11 co11ce11trate ) ~R a11d recovery 25 230 3 0.06 0.28 gai11 = i111proved recovery (IR)curre11t recovery (CR) (%) 26 240 4 0.08 0.22 Most of tl1e 5 para111eters above are variables, witl1 so111e fluctu ati11g 111ore tl1a11 otl1ers. OT a11d OD are relatively co11sta11t i11 a11y well 111a11aged operatio11s. AG is reaso11ably 1<11ow11 tl1rougl1 good Tl1e auto111ated 111i11eralogy worl< is esti111ated to 11ave brougl1t geo-111etallurgical worl<. MP is co11trolled by 111arl<et forces, but forward tl1e i111ple111e11tatio11 of fi11e gri11di11g project by, say two ca11 be esti111ated. is critical a11d tl1e 111ost u11certai11 para111eter. ~R years. Delayi11g tl1e project by two years would 11ave resulted i11 by defi11itio11 sl1ould be 11igl1er tl1a11 zero a11d sl1ould be less ~R. a11 opportu11ity loss of arou11d US$53 111illio11 (based 011 tl1e i11co111e tl1a11 100 - CR. So, M usually 11as a wide variatio11. 111 tl1e case of US$75 111illio11 per year, 111i11us US$15 111illio11 per year ru1111i11g study above, M tur11ed out to be 5%. However, tl1is value 111ust costs a11d tl1e cost of i11vesti11g US$250 111illio11 two years earlier). be esti111ated wl1e11 tl1e process cl1a11ge is proposed. 111 Al1glo Y. Gu et al. / Minerals Engineering 58 (2014) 100-103 103 Plati11u111's case, tl1e esti111ated M value at tl1e ti111e of proposal 111al<i11g. Witl1out tl1e i11for111atio11, a decisio11 ca11 110t be 111ade was 2%. Metallurgists were co11fide11t tl1at 2% sl1ould be acl1ievable a11d tl1e opportu11ity is lost. Tl1e loss is tl1e value of i11for111atio11. a11d tl1at tl1e retur11 is large e11ougl1 to justify tl1e process cl1a11ge. Tl1e retur11 011 i11vest111e11t for auto111ated 111i11eralogy is particularly Let us use a l1ypotl1etic processi11g pla11t of 111oderate size to l1igl1 at tl1e start of a11 opti111isatio11 project. de111011strate l1ow to 111odel tl1e retur11 wl1e11 we l1ave two or 111ore Tl1is is tl1e first ti111e a11 atte111pt l1as bee11111ade to qua11titatively variables. Tl1e curre11t lo11g ter111 average recove1y for tl1is pla11t is value tl1e co11tributio11 of auto111ated 111i11eralogy for 111i11eral pro 85% a11d expected i111proved recovery after process 111odificatio11 cessi11g pla11t opti111isatio11. Tl1e above 111etl1od ca11 be used to esti- is 8 7%. Tl1e basic i11for111atio11 witl1 calculated retur11 is listed i11 111ate tl1e value of 111etallurgical test worl< i11 tl1e co11text of eitl1er Table 2. ore source cl1aracterisatio11 (geo111etallurgy) or pla11t opti111isatio11. I11 Table 2, tl1e 111ea11 retur11 value is calculated fro111 tl1e 111ea11 Tl1e 111etl1odology developed i11 tl1is paper ca11 be applied to dif values of M a11d MP a11d so fortl1 for low a11d l1igl1 values. Tl1ese fere11t co111111odities a11d differe11t scales of 111i11i11g operatio11s. Eacl1 values represe11t a wide ra11ge of possibilities. If we 11eed to i11vest, 111i11e will l1ave differe11t operatio11al para111eters, sucl1 as to1111ages, say US$50 111illio11 to 111odify tl1e circuit i11 order to obtai11 tl1e grades of ore a11d tl1e price of tl1eir products. It is ofte11 tl1e case tl1at esti111ated recove1y i11crease tl1ere is a ra11ge of possible retur11s. tl1e 111etallurgist 11eeds to evaluate differe11t pla11t 111odificatio11 op However, we do 110t l<11ow tl1e lil<elil1ood associated witl1 eacl1 tio11s. Tl1e si111ulated results for eacl1 of tl1e optio11s ca11 be used to possible outco111e, a11d tl1erefore use a Mo11te Carlo si111ulatiorr to ra11l< tl1e111 so as to allow for tl1e selectio11 of tl1e optio11 tl1at ca11 provide detailed i11for111atio11 to support tl1e decisio11. deliver tl1e best possible outco111e witl1 acceptable risl< profile. Give11 tl1e basic i11for111atio11 i11 tl1e above table, a si111ulatio11 of Tl1e 111odel developed l1ere is by 110 111ea11s tl1e 111ost co111prel1e11- 5000 sce11arios produces tl1e distributio11 of retur11s as sl1ow11 i11 sive 011e. However, it sl1ould be 111ore tl1a11 adequate to assist tl1e tl1e Table 3 a11d Fig. 1. pla11t 111a11agers wl1e11 tl1ey co11sider pla11t opti111isatio11 a11d i11vest- Table 3 sl1ows tl1at tl1ere is a 36.5 % cl1a11ce tl1at tl1is i11vest111e11t 111e11t i11 auto111ated 111i11eralogy. will produce a retur11 of US$100 111illio11 or 111ore per year. It also Tl1e 111odel is 110t a blacl< box. It depe11ds 011 tl1e esti111ates of sl1ows tl1at tl1ere is a 19.5% cl1a11ce tl1at tl1e retur11 will be well be u11certai11ties of tl1e para111eters (tl1eir ra11ges -co11fide11ce i11ter low US$50 111illio11 per year a11d a 2.3% cl1a11ce of 110 retur11 at all. vals). Hubbard also provides ways for users to calibrate tl1eir esti- Witl1 tl1e detailed i11for111atio11, pla11t 111etallurgists a11d 111a11ag 111ates, so tl1at tl1ey ge11erate realistic values. ers ca11 select betwee11 differe11t i11vest111e11t strategies a11d tl1e level of i11vest111e11t a11d calculate tl1e retur11 011 i11vest111e11t (ROI) for Acl<nowledgements eacl1 of tl1e optio11s witl1 a certai11 degree of co11fide11ce. Witl1 tl1at i11for111atio11, tl1ey ca11 pla11 a11d carry out 111ore auto111ated 111i11eral Tl1e first autl1or would lil<e to tl1a11l< FEI for tl1eir fi11a11cial sup ogy a11d 111etallurgical test worl< to reduce tl1e u11certai11ties a11d i11- port to a researcl1 project, resulti11g i11 tl1is paper. We would lil<e to crease tl1e co11fide11ce i11 tl1e M esti111ates, tl1ereby si111ulati11g 11ew tl1a11l< colleagues at JI<MRC for sti111ulati11g discussio11s a11d co111- retur11 distributio11s. Tl1is iterative process ca11 co11ti11ue u11til a 111e11ts. We also express our appreciatio11 to tl1e a11011y111ous clear decisio11 ca11 be arrived at. I11 eacl1 iteratio11 tl1e co11tributio11 reviewers for tl1eir co11structive co111111e11ts. fro111 auto111ated 111i11eralogy is equivale11t to tl1e value of reduced risl<, sa111e as for 111etallurgical test worl<. References As 111e11tio11ed before, tl1e retur11s 111odelled above are a co111- Hubbard, D.W., 2010. How to Measure Anything: Find the Value of"lntangibles" in bi11ed value of auto111ated 111i11eralogy a11d 111etallurgical test worl<. Business. second ed. john Wiley & Sons, ISBN 978-0-470-53939-2. As tl1ese two discipli11es are i11terrelated a11d co111ple111e11ta1y, tl1e Lotter, N.O., 2011. Modern process 1nineralogy: an integrated 1nulti-disciplined relative co11tributio11 of tl1ose two parts of worl< is deter111i11ed lar approach to flowsheeting. Miner. Eng. 24, 1229-1237 gely by esti111ati11g tl1e level of risl< reduced by eacl1 activity. Good Lotter, N.O., Di Feo, A., l<or1nas, LJ., Frago1neni, D., Co1neau, G., 2010. Design and 1neasure1nent of s1nall recovery gains - a case study at Raglan concentrator. esti111atio11 de111a11ds i11sigl1t l<11owledge a11d experie11ce witl1 accu- Miner. Eng. 23, 567-577. 111ulated case studies. MacDonald, M., Adair, B., Bradshaw, D., Dunn, M., Latti, D., 2011. Learnings fro1n five years of on-site MLA at l<ennecott Utah copper corporation (1nyth busters through quantitative evidence). In: Proceedings of the 10th International 5. Conclusions Congress for Applied Mineralogy, Trondhei1n. Rule, C.M., 2011. Stirred 1nilling - new co1n1ninution technology in the PGM A 111odel is developed usi11g Mo11te Carlo si111ulatio11 allowi11g indust1y. In: SAIMM conference platinu1n in transition, boo1n or bust. SAIMMJ., 71-78. tl1e i11put of 111ultiple variable para111eters to calculate tl1e probabil Rule, C.M., Schouwstra, R.P., 2011. Process 1nineralogy delivering significant value at ity of retur11. Tl1e value of auto111ated 111i11eralogy is esse11tially tl1e A11glo Platinu1n concentrator operations. In: Proceedings of the 10th value of i11for111atio11 tl1at it provides to allow i11for111ed decisio11 International Congress for Applied Mineralogy, Trondhei1n. 1 Monte Carlo sin1ulations or Monte Carlo experin1ents are a broad class of con1putational algorithn1s that rely on repeated randon1 san1pling to obtain nun1erical results. The n1ethod is used in a wide range of fields fron1 econon1ics to nuclear physics. In this paper, we use it to n1odel the variation of all the variables influence the return of investn1ent.