Challenges for LCAs of Complex Systems: T he Case of a Large-Scale Precous Metal Refnery Plant Anna Stamp, Chrstna E.M. Meskers, Markus Remer, Patrck Wäger, Hans-Jörg Althaus and Roland W. Scholz Abstract Umcore Precous Metal Refnng (UPMR) runs a hgh-tech ndustral metal refnery whch recovers 17 dfferent metals from end-of-lfe consumer products and from by-products of the non-ferrous ndustry. We present an approach for an attrbutve gate-to-gate LCA study of ths system, whch s charactersed by mult-nput/mult-output processes, changng feed compostons and tme lags. We propose fve assumptons to reduce the complexty of the hghly dynamc system. We compled nventory data for over thrty sub-processes and allocated t over the metals passng the sub-process by ether a mass-based or metal revenue based allocaton. The exemplary results for rhodum, platnum, tellurum and copper (mpact assessment method: global warmng potental) show a hgh dependence of allocaton choce and dfferent patterns of the metals for metal revenue based allocaton due to the hgh volatlty of prces. 1 Introducton Fosterng the use of more sustanable products can help to reduce the envronmental footprnt of our daly lvng. Lfe cycle assessment (LCA) studes can support the development of such products by quantfyng envronmental mpacts of a product over ts full lfe cycle. The lfe cycle ncludes resource provson, manufacturng and use, as well as end-of-lfe treatment, whch can ether comprse waste management or reuse respectvely recyclng. The supply of metals as part of the resource provson n a product's lfe cycle ncludes the refnng of ether prmary materal or secondary materals (producton scrap and end-of-lfe products). In ths contrbuton we present an approach for performng an LCA study of the process of a large metal refnng company, Umcore Precous A. Stamp ( ) Swss Federal Laboratores for Materals Scence and Technology, Dübendorf, Swtzerland ETH Zurch, Zurch, Swtzerland e-mal: anna.stamp@empa.ch C.E.M. Meskers Umcore Precous Metals Refnng, Hoboken, Belgum M. Remer P. Wäger H.-J. Althaus Swss Federal Laboratores for Materals Scence and Technology, Dübendorf, Swtzerland R.W. Scholz ETH Zurch, Zurch, Swtzerland M. Fnkbener (ed.), Towards Lfe Cycle Sustanablty Management, 247 DOI 10.1007/978-94-007-1899-9_24, Sprnger Scence+Busness Meda B.V. 2011
248 Anna Stamp et al. Metals Refnng (UPMR), whch produces - amongst others - precous metals (e.g. platnum group metals) and specal metals (e.g. ndum and tellurum) from endof-lfe consumer products and from by-products of the non-ferrous ndustry. A major challenge of ths project was to master the fact that the lfe cycles of these metals are charactersed by strong nterlnkages of metal streams. Ths holds true for the prmary producton, as these metals often are by-products of major metals, but also for secondary producton snce these metals typcally occur n complex mxtures n end-of-lfe products [1]. Hence, metal refnng mostly nvolves several metals at a tme and requres effcent separaton and refnng processes to both close the resource cycle and mnmse envronmental mpacts early n the supply chan. 2 Methodology 2.1 Goal and scope 2.1.1 Objectve of the study We present an approach to quantfy the envronmental mpacts of 17 metals refned at the faclty of UPMR n Hoboken, Belgum, from a gate-to-gate perspectve. The Hoboken system conssts of strongly nterlnked sub-processes (n partcular smelter, blast furnace and refnery steps, see Fgure 1), whch have a mult-nput/mult-output character, nternal loops, changng feed compostons and tme lags. Ths paper ams at ) explanng the overall approach, whch led to an Excel tool that can be lnked to the exstng montorng system of the company, and ) hghlghtng how methodologcal choces, such as allocaton rules, affect the results. We do not focus on ndvdual lfe cycle nventores of specfc metals, but show exemplary results for the metals platnum, rhodum, tellurum and copper for one mpact assessment method. Further results wll be shown n a forthcomng publcaton. The approach developed n ths project can serve as a frst step towards a comparatve study wth other prmary or secondary producers.
LCM of Processes and Organsatons 249 Fg. 1: The Hoboken metal refnery system (adapted from [2]). The trangle ndcates the man entry pont for feed materal. Note that the number of sub-processes and arrows are not complete. 2.1.2 System boundares and data sources System boundares for the foreground system modelled n ths study were defned by the Hoboken ste,.e. all process steps that are related to metal recovery at the Hoboken plant were ncluded, except samplng and assayng (the analytcs necessary before a feed materal s accepted by UPMR). The two non-metal products of the system, sulfurc acd and slag (sold to cement producers), are excluded due to ther mnor mportance (cut-off). The process flow chart (see Fgure 1) was broken down nto over thrty sub-processes for whch data on materal and energy nputs was avalable from the nternal reportng system of the company. Data on outputs, that s emssons and waste streams, was only avalable on a coarser scale, therefore these measurements had to be reallocated to the subprocesses based on the knowledge of the responsble process engneers. The annual metal output per sub-process s montored by UPMR and nternal materal loops were calculated based on transfer coeffcents avalable from UPMR's metal flow model. The gate-to-gate standpont of ths study mples that the feed materal, whch could be secondary materal as well as valuable by-products of prmary metal producton, was not accounted for. In other words: the feed was
250 Anna Stamp et al. assumed to be free of envronmental mpacts. We accounted for the metal product qualty as t leaves the Hoboken ste n a marketable product. As Table 1 shows, ths can ether be a pure refned metal or an ntermedate product that can be further refned elsewhere. The background system (e.g. the supply chans for materals and energy provson) was modelled by usng the Econvent database v.2.2 [3]. For some nternal processes takng place at the Hoboken ste, whch provde auxlary process nputs, we compled own nventores. Tab. 1: Metals processed at Hoboken and grade of the product leavng the plant ( + ntermedate products, ether further refned elsewhere or drectly sold on the market; *produced as techncal grade and hgh purty products, for the analyss t was theoretcally assumed that all s techncal grade; ** leaves the plant as refned product and as an alloy wth B; ++ not avalable) Metal Grade [%] Metal Grade [%] Precous metals 1. Platnum (Pt) 99.95 5. Ruthenum (Ru) + 2. Palladum (Pd) 99.95 6.Gold (Au) 99.99 3. Rhodum (Rh) 99.95 7. Slver (Ag) 99.99 4. Irdum (Ir) + Specal metals 8. Antmony (Sb) - sodum + 11. Selenum (Se)* heyhydroxo antmonte 99.50 9. Bsmuth (B) - n Pb + 12.Tellurum (Te) alloy 99.50 10. Indum (In) 99.99 Other metals 13. Lead (Pb)** 99.97 15: Tn (Sn) - calcum stannate + 14. Copper (Cu) cathode NA ++ 16. Nckel (N) + 17. Arsenc (As) + 2.1.3 Functonal unts The functonal unts (FU) of ths study refer to 1 klogram specfc metal content leavng the Hoboken system, n a grade and specfcaton as ndcated n Table 1. Our approach only accounts for metals n the system, whle other materal flows, such as those formng compounds, are neglected.
LCM of Processes and Organsatons 251 2.1.4 Allocaton rules Insde the Hoboken system, allocaton rules need to be defned snce the metals take dfferent routes through the system and occur n dfferent combnatons n each sub-process. We present results for dfferent allocaton ratonales that are based on: Mass (klogram),.e. physcal allocaton Value (US Dollar) of fully refned metal,.e. revenue/prce allocaton, wth o average prces between 2000 and 2010 o average prces n 2000 o average prces n 2010 For the revenue allocaton ("revenue" = prce of the metal tmes ts amount), we used the prces of the "pure" metal, even though some metals leave the plant as ntermedate product (see Table 1), as the prces of the ntermedates were not possble to determne. Snce metal prces showed hgh volatlty n recent years, we tested the robustness of revenue allocaton by calculatng t wth ) a ten year average, ) wth the prce before the metal boom and ) wth the prce durng/after the metal boom. The results can be further valdated by calculatng all allocatons for two years; wth mass flow data from 2008 and 2009. 2.2 Lfe cycle nventores (LCI) 2.2.1 Quantfcaton of nventory data per sub-process In ths frst step, an nventory for each sub-process n the foreground system was establshed. Inventory data ncludes the use of materals and energy and the dscharge of waste and emssons. The nternal reportng system of UPMR accounts for the use of energy and auxlary materals per process unt. In our study we calculated wth over thrty sub-processes. Data on emssons and waste streams was avalable on a coarser scale, for nstance when several processes use the same chmney, and had to be allocated to the sub-processes usng a combnaton of judgments by engneers and cost allocaton models. For nstance, emssons to water were allocated to waste water streams that are pped to the waste water treatment plant (WWTP) based on judgments of the
252 Anna Stamp et al. process engneers. The waste water streams, n turn, were allocated to subprocesses based on an nternal cost allocaton model. Ths cost allocaton model was also the bass to dstrbute materal and energy nputs as well as outputs (emssons and waste) of the WWTP to the sub-processes. For some nternally produced materals (dfferent process water qualtes, pressursed ar and steam) we compled nventores for producng one unt based on data from the plants provdng the materal. These plants take advantage from waste heat produced elsewhere n the Hoboken system, makng the producton process more economc. For sulphurc acd that s produced from SO 2 n the smelter off gas, we dd not provde an own nventory, snce ) only a small fracton s used nternally and most sulphurc acd s sold on the market and ) the producton process does not devate from ndustral standards. 2.2.2 Dstrbuton of nventory data over metals n sub-process In the second step, the nventory data of the sub-processes was allocated to the metals passng that sub-process. The number of metals usng a sub-process can dffer: Whle the smelter s part of every recovery process as all metals pass through t, sub-processes n the precous metals refnery for nstance can be part of only one or few metal recovery processes. Snce the metals' retenton tmes n the system and ther annual share n the feed vary strongly, one major challenge of ths project was to answer the queston, how annually reported nventory data can be allocated to sngle metals n a sub-process. In our approach we reduced the complexty of the hghly dynamc system by applyng the followng three key assumptons: 1) The use of energy and auxlary materal depends on the quantty of metal handled n the sub-process, whle dfferences n the type of metal (ther "qualty"), are neglected (for nstance could some metals be more demandng to be treated n a process than others); 2) The amount of materal and energy needed per sub-process and ton of metal handled stays constant over tme; 3) Each process s short n tme. As a consequence, we frst looked at each sub-process n separately and dstrbuted the nventory data m over the metals handled n that sub-process based on ther share on the total metal stream (Equaton 1, see Fgure 2 for explanaton of terms). In flown, Inv n, m, Invn, m (1) In flow n, tot
LCM of Processes and Organsatons 253 The annual nput flow of metal n sub-process n (In-flow_n,) s calculated by dvdng the annual output of metal, whch s regularly reported for each subprocess n (Out-flow_n,), by ts yeld of the whole process (Equaton 2). We use data on nputs nstead of outputs n order to account for nternal losses. Out flown, In flown, (2) yeld n, The total metal nput flow for the sub-process n (In-flow_n,tot) s the sum of the flows of each metal (Equaton 3) 17 In flow In flow (3) n, tot 1 For the metal revenue based allocaton, the physcal metal flows were translated nto a money flow by multplyng them wth ther prce per mass unt. The metal stream ncludes nternal loops (the metals can pass a sub-process several tmes). n, Fg. 2: Schematc vew on one sub-process. A sub-process handles one to all metals onto whch ts use of auxlary materals and energy and ts generaton of waste and emssons need to be allocated. 2.2.3 Complaton of lfe cycle nventores per metal In the thrd step, an nventory per metal for the whole Hoboken plant (Inv_Hoboken,m,) was calculated. The nventory data per sub-process (Inv_n,m,) was dvded over the metals gong through that sub-process accordng to ther share on total metal stream, whch ncluded nternal loops (see Equaton 1). In order to comple a complete nventory for each metal n the whole Hoboken plant for the functonal unt of one klogram specfc metal content n the product, the reference of each sub-process needed to be changed to the metal nput excludng nternal loops. The resultng nventores for metal per sub-process could then be summed up and form the complete nventory of metal n the Hoboken plant.
254 Anna Stamp et al. A metal's throughput generated n sub-process n by the nput of 1 klogram of the metal n the smelter was known from UPMR's "metal studes", whch are nternally prepared models on metal flows that are based on measurements of how the outflow of a metal from a sub-process dvdes over further sub-processes. Wth these "splt factors" (e.g. 90% of metal leaves the smelter to sub-process n1 and 10% to sub-process n2) for every metal and every sub-process t was possble to model how many tmes a metal n the annual feed mx enterng at the smelter crculates n the plant before t leaves Hoboken. The calculaton of the total Lfe Cycle Inventory of 1 klogram metal n the Hoboken plant s shown n Equaton 4. Inv In flow 34 n, Hoboken, m, nvn, m, (4) n 1 In flown smelter, Ths proceedng mples that two more assumptons have to be added to the three mentoned n chapter 2.2.2: 4) When calculatng the number nternal loops for one metal n one subprocess, we used a theoretcal model that s based on the flow of the annual feed composton through the Hoboken system, assumng that all nput materal enters at the smelter. In realty, feed materal that s already smlar to the materal mx of another sub-process can enter later n the process, whch s not possble to account for n our approach. Assumng the smelter to be the sngle entry pont s close to realty, but neglects that some feed s treated more effcently. 5) We used the average annual feed composton when calculatng the number of nternal loops; however, snce the retenton tme n the system vares per metal as well as ther annual share n feed materal, metals that enter the system n the last month of year y wll leave the system at some moment n year y + 1. Ths mples that we assume that the share of a metal n the feed (whch can vary between years) does not affect the numbers of ts nternal loops. 2.3 Integraton wth lfe cycle mpact assessment (LCIA) By followng the steps outlned before, we obtaned a lst of nputs (materal and energy) and outputs (emssons and waste), whch were related to one klogram of metal refned n the Hoboken plant. The envronmental mpacts of these background materals and servces were quantfed by usng the Econvent database (v2.2) as mplemented n Smapro.
LCM of Processes and Organsatons 255 We chose dfferent mpact assessment methods but focus here on the global warmng potental, applyng a tme frame of 100 years (GWP) [4]. Results for other mpact assessment methods wll be publshed n due tme. 3 Results The choce of the allocaton ratonale determnes how much envronmental mpact s owed to one klogram of a specfc metal present n the Hoboken plant. Ths s llustrated n Fgure 3 for the mpact assessment method GWP and four selected metals: Platnum, the probably best known and most used platnum group metal; Rhodum, the most expensve of all metals refned n Hoboken; Tellurum, used n thn-flm photovoltac; Copper, one of the carrer metals, wth a relatvely low prce, present n hgh quanttes whle t s not the man focus of UPMR's busness. Fg. 3: Results for materal flow data from 2009. The bars refer to dfferent allocaton methods (black: mass allocaton; grey tones: revenue allocaton, wth average prce of metal between 2000-2010, average prce of metal n year 2000 and average prce of metal n year 2010). The values are gven as percentage of the results for rhodum, mass allocaton (n logarthmc scale). Note that the results presented here are prelmnary, snce they do not nclude sold waste and a last check of some numbers s lackng. The decson on allocaton ratonales has a sgnfcant mpact on the results of each metal, as Fgure 3 shows. Dependng on a metal's prce, the mpacts can
256 Anna Stamp et al. ether be sgnfcantly hgher (e.g. by more than an order of magntude for rhodum) or lower (e.g. for copper) compared to a purely mass based allocaton. Rhodum s the most expensve precous metal wth an average prce of about 85'000$ per klogram between 2000 and 2010, whch s by several orders of magntude hgher than the prce for a klogram of copper (average prce between 2000 and 2010: 4.25$ per klogram). Platnum and tellurum are n-between wth an average prce of about 32'000$ and 140$ per klogram, respectvely. The three metal revenue based allocatons depcted n Fgure 3 show dfferent patterns for each metal, whle the rankng remans stable for the four metals chosen. In the frst decade of the 21st century, the metal market experenced hgh volatlty, whch, however, dd not affect all metals smlarly. Rhodum, for nstance, had a prce of about 17'000$ per klogram n 2003 and of 211'000$ n 2008, whch dropped down to 79'000$ n 2010. The prce curve for platnum s smoother, whch s reflected by the smaller varatons between the revenue allocatons. However, the senstvty towards choosng a reference year for metal prces can not only be explaned by the volatlty of the prce of the respectve metal. In addton, the share of the value of a specfc metal n the total value of metals usng the same process s relevant. Therefore, the effect of consderable prce dfferences dmnshes later n the refnery steps, where only few metals are stll nvolved n the same process. 4 Dscusson We lsted fve assumptons underlyng our approach n secton 2.2.2 and 2.2.3. The frst assumpton referrng to dfferences n a metal's qualty s probably the most relevant. Includng rather than omttng ths aspect would mply to look at thermodynamc characterstcs of the metals and ther compounds and probably also to dstngush ther composton n the feed. However, due to the hgh resoluton of more than thrty sub-processes consdered n our approach, we can capture the dfferent nvestments for refnng specfc metals by ) only allocatng process mpacts to the metals actually demandng t, and, related to that, ) accountng separately for the last refnng steps of most metals (whch are n general the most "expensve") and ) ncludng the amount of nternal loops of a metal. The second assumpton mples that there s no "base load" of the subprocesses; however t s reasonable to assume that UPMR s tryng to keep the load constant for a hgh effcency, whch can be valdated by cross-checkng wth dfferent years (provded that no nstallaton of new sub-processes has changed the process flow chart ). The second assumpton together wth the thrd s the bass for
LCM of Processes and Organsatons 257 our approach that avods a detaled representaton of the hghly dynamc character of the process by focusng on sub-processes, where the factor tme s less relevant (the factor tme ncludes dfferent retenton tmes from several weeks to a number of months and changes n annual feed composton). The exemplary results for four metals and one mpact assessment method showed that choosng between a purely mass-based allocaton and a metal revenue-based allocaton has a strong effect on how the mpacts of the large refnery plant of UPMR are dstrbuted over the metals. The ratonale of the mass-based allocaton s that all metals refned at Hoboken are consdered equally mportant as drvers for the process. For the metal revenue allocaton, the ratonale s that the process s drven by the more economcally valuable elements, whch s affected by the volatlty of metal prces. However, a short term reacton on fluctuatons n metal prces s hardly possble for UPMR, snce ) the feed spends a certan tme n the system, ) feed materal s secured for longer tme n the future, ) metal prces are hedged (.e. fxed when the materal arrves at the plant, for payment n the future), and v) the large nfrastructure can only be adapted n the medum and long term. When consderng UPMR s busness model n more detal, t appears that both allocaton ratonales cannot fully account for ts specfc ncentves: In metals refnng dfferent metals have to be handled at the same tme because they are fed to the plant as metals mxtures n the dfferent feed materals. Ths means that not only the metals as such drve the process, but also the type of feed, whch ranges from by-products of the non-ferrous ndustry to e-scrap and automotve catalysts. The busness model of UPMR s to offer clents to refne metals from feed they provde. The clent receves the refned metal and pays Umcore a treatment fee for ths servce. From ths pont of vew, all metals can n theory equally contrbute to the company's revenue. In that sense, treatment fees are more relevant as drver of the process than metal prces. Furthermore, some nterdependences n the process are not consdered by the allocaton ratonales, as some metals fulfl a functon as carrer metal n the process, whch makes ther presence relevant for recovery processes of other metals. 5 Concluson and outlook Our study provdes nsghts nto envronmental mpacts early n the product supply chan by focusng on the process of a large scale, ndustralsed metal refnery. In partcular, t shows how 17 metals enterng the same refnery plant dffer n ther envronmental mpacts, and how senstve the analyss s towards
258 Anna Stamp et al. mass-based versus metal revenue based allocaton. When nterpretng the results of ths study, t has to be kept n mnd that they are only vald for the specfc process desgn and product portfolo of UPMR, that s: the process s optmsed for a feed mx and not for a specfc metal. Furthermore, the study only refers to a specfc average annual feed composton at UPMR and cannot be generalsed for prmary and secondary sources (as both types of feed enter the process). The current work shows a large number of academc and practcal challenges that are assocated wth obtanng relable LCA data for large, complex metallurgcal operatons. Usng LCA as a tool to quantfy the performance of recyclng processes s mportant and necessary to move towards a sustanable socety. Nevertheless, t needs to be kept n mnd that the qualty of the nventory data has to be good, and the lmtatons of the model have to be recognsed to make wellthought decsons n the operatonal and polcy area. For future work t s necessary to nclude further specfc characterstcs and dynamcs of the complex metallurgcal processes n the approach. More specfcally, t would be helpful to nclude a dfferentaton of nput materal (and e.g. take nto account dfferent heatng values), as well as dstngushng between the metals produced from a specfc feed. One opton would be to change the research queston to whch envronmental burdens and benefts occur from treatng a certan feed, wth e.g. CO 2 beng a burden and platnum recovery beng a beneft. In that case t would also be possble to allocate envronmental burdens of the treatment process to the lfe cycle of producng the feed materal. References [1] Hagelüken C, Meskers CEM (2010) Complex lfe cycles of precous and specal metals. In: Graedel T, Van der Voet E (eds) Lnkages of Sustanablty, 1st edn. MIT Press [2] <http://www.precousmetals.umcore.com/pmr/process/> (accessed 14.04.2010) [3] <http://www.econvent.ch/> (accessed 14.04.2010) [4] IPCC (2007) IPPC Clmate Change 2007: Synthess Report, Cambrdge Unversty Press