r^ %:$»*

Process simulation for waste managementPTUCKER, BA, PhD, MlnstP, CPhys and KA LEWIS, BScWarren Spring Laboratory, Stevenage, UK

SYNOPSIS Acritical assessment of available options for waste separation, recycling and energy recovery is anessential strategic step in formulating an economic and technically viable waste management policy TTieunderpinning data for each assessment includes the provision ofcharacterisations ofthe waste and quantificationof its response to separation (either centrally or at source). These data are however difficult and expensive toobtain (at least conventionally through survey and practical work). They can alternatively be derived to thenecessary resolution but very much more cost-effectively, using predictive modelling methods, (i) Specific wastecharacterisation data can be predicted from alternative data sets, once these are formulated as functions ofmeasurable objective criteria, (ii) Generic separation models based on measured performance data can bedeveloped to provide the required catalogue of separation responses, for the whole spectrum of separationregimes. This paper describes the integrated role of 'characterisation models' and 'process* models This ispresented through astepwise approach outlining the technology that has been developed and the technology thatstill needs to be developed to complete the overall system.

NOTATION

JlJzJn

Ba.,BI

K,OV\,OV\P

slts

power index of breakagefunction

process feed mass flow rate *index variables for size classes, i*<:iindex variable for physicalproperty classes l,2...nindex variable for process outputstream

process product mass flow ratebreakage functionidentity matrixconstant in breakage functiondevice operating variablesset of model parametersselection functions

transfer coefficient

size class intervals

1 INTRODUCTION

The development of processes for the recovery ofmaterials and energy from waste involves significantamounts of practical testwork to establish feasibilities,

zm&mr

to provide assurances that performance targets will bemet and to optimise performance. Such testwork hasoften been neglected due to thehigh cost involved andhas led to cases where full scale plant failed to meetexpectations orspecification. Further, the full environmental impact of the developed process has oftenproved difficult to predict.

A robust simulation model for waste treatment canaid the planning, design and operation of wasteprocessing operations by providing a substantial partof the underpinning and necessary information,significantly reducing testwork requirements. Themodel can allow the conceptual design of plant, highlight uncertainties and sensitivities in the processflowline or flowsheet, provide comparison of processoptions and help trace pollutant pathways. It can alsoenable plant sizing, quantification of the effects ofoperational factors such as source separation, feedvariability and equipment set-up and can furtherallow optimisation ofthe process against any objectivecriterion (such as maximising energy recovery). Moregenerally, the modelling approach will help to reducedevelopment risk.

Simulation models are now well proven in thechemical, paper and minerals industries but have, asyet, had little impact inthewaste processing, although

47

some pioneering work has been carried out, mainly inthe United States.

An important precursor to process simulation istheestablishment of the composition ofthe process feed,or proposed feeds to the process, at the required levelof detail. This can become expensive in terms of theanalyses needed. It may also be impossible to achieveIogistically orwith adequate representivity or in time.A means of estimating the necessary dataindependently, perhaps by utilising other data sources,then becomes essential. This estimation facility iscurrently being developed under the NationalDomestic Waste Analysis Programme. Thecomplementarity of the national waste statistics andthe process simulation method will provide anintegrated tool which has many applications in wasteand energy management. This integrated concept isshown schematically in figure 1. The concept isdeveloped in this paper by stepping progressivelythrough the technology involved.

The categories have been selected based on theexperience and practicalities of sorting to providegeneration rates for all the major recyclablecomponents within the waste. The analysis itself isundertaken by hand-sorting a sample of thewaste andscreening each component by size. Practicalconsiderations and problems ofsample representivityconstrain the size ofsample that must be analysed. Innormal circumstances, a full refuse disposal vehicleload of approximately 5 tonnes is treated. A 5%relative error on the analyses of most of the majorcomponents can usually be achieved, with this size ofsample.

Chemical analysis of the whole sample, or of thesub-fractions, provides further information and detailon the composition of the waste. Determinations ofthe moisture content, calorific value, the containedvolatile matter and fixed carbon ash levels provide theessential information on the 'energy value of thewaste'. Analyses for heavy metals and for other keyelemental components provide information on thepotential toxicity of the waste. Further analyses forother toxins such as PCBs and dioxins are added ifrequired.

Characterisation of waste to this level of detail isthe minimum requirement for process simulation. Italso defines the requirement for collecting the UKnational statistics on the waste composition arisingfrom households.

2 WASTE CHARACTERISATION

A basis for domestic (dustbin) waste characterisationis provided by analysis of the waste by materialcomponent and size distribution. In the UK, six sizefractions are generally used for this categorisation:+ 160 mm., -160+80 mm., -80+40 mm., -40+20mm., -20+10 mm. and -10 mm. Eleven mainmaterial categories (table 1)are usedbut when a moredetailed analysis is required, these fractions arefurther divided into thirty three sub-categories.

Table 1 Material categories of domesticrefuse i

Categories

1 Paper & card2 Plastic film

3 Dense plastic4 Textiles

5 Miscellaneous combustibles6 Miscellaneous non-combustibles7 Glass

8 Putrescibles9 Ferrous Metal

10 Non-ferrous Metal11 -10 mm. fines

48

3 THE NATIONAL WASTE ANALYSISDATABASE

The National Domestic Waste Analysis Programme(1) was set up by the UK Department of theEnvironment as a means of improving the UK's wastearising statistics, validating these data and developinga method of monitoring trends in waste arisings on acontinuous basis. In the programme, a rationale wasdeveloped which will allow the correlation ofdomesticwaste generation rates and waste composition (asdefined above) with source details. These sourcedetails are currently based primarily on socioeconomic information. The basis of this rationale willbe expanded below. The Programme seeks tobuild upthe base UK waste generation statistics through fullcharacterisation of some20 keyselected wastestreamseach year over the Programme's duration. The datawill provide a centralised, national reference facility,developed and held as a computer database at WarrenSpring Laboratory.

ii

V

c

a

ft

1c

d(

3.

a

ar

re,

pu

w<

pu

Fo

20th<

A(

pn

3.1 Source characterisation

The National Waste Analysis Database uses theACORN classification system to provide the primarybasis for source categorisation. Developed byCACI,ACORN (A Classification of ResidentialNeighbourhoods) is a geodemographic classificationsystem which allows census data to be classified intovarying socio-economic types, according to thesortofresidential area people live in.

An underlying assumption, of the Programme, isthat people who live in similar neighbourhoods havesimilar purchasing and lifestyle habits andconsequently will generate similar wastes. Forexample, low income households are low wastegenerators mainly because they purchase less and somere is little packaging etc. to dispose of. Highincome households are better placed to purchaseluxury goods such as pre-packaged food stuffs, wineand magazines, which generate large quantities ofwaste high in paper and glass content. Althoughhousehold socio-economic profile is not the onlydetermining factor in wastegeneration, it hasprovideda credibleand defendable startingpoint for the study.Other contributory factors affectingwastecompositionwill be more fully detailed as the study progresses.These factors will include seasonal and regionaldifferences, collection methods and the use of coalfires etc.

The waste arising and composition data, collectedin the study, has been organised within the NationalWaste Analysis Database according to ACORNcategory(table 2), with each data set relatingsolelytoa single ACORN source group. Each data set isfurther tagged with its date of collection, collectionlocality and collection method to allow for futuredevelopments.

3.2 Statistical models of waste composition

On the above rationale, the waste composition andarisings for any area (eg. enumeration district, cityorregion) can be estimated, to a first approximation,purely from the ACORN profile of that area. Thewaste composition will be compounded from the basicpure ACORN building blocks, stored in the database.For example, given an ACORN profile of 30% B,20% D, 40% F and 10% J, the waste composition forthat area is estimated by combining data from pureACORN groups B, D, F, J in the same percentageproportions.

As more data is accumulated on measured waste

Table 2 The ACORN classification system

ACORN Housing Type

A Agricultural villages and farmsB Modern family housing, young familiesC Older housing, intermediate statusD Older terraced housingE Quality council estates, higher incomesF Council estates, older peopleG Council estates, inner cities, low incomeH Mixed inner metropolitan areasI High status non-family housingJ Affluent suburban housingK Better-off retirement areas

compositions from a wider range of UK sources, therelative influences of all the contributory factorsaffecting waste generation will become betterquantified. This will enable the creation of newstatistical models and corrections to account, forexample, for seasonal variations in weight arisings(eg. at holiday resorts), seasonal variations in wastecompositions (eg. garden refuse content), regionaldifferences (if any) between towns having similarACORN profiles, and changes in consumer habit ifthe collection method is changed.

Although at an early stage of the overallprogramme, the base statistics and their manipulationthrough the simple models described above havealready started to give assistance to Local Authorities:to more closely define waste arisings, to developwaste disposal and recycling plans and to improvecontractual arrangements for waste managementservices. The evaluations of these options are stillcarried out off-line. The next, and important, step isto use the power of computer models and processsimulation techniques to assist in the evaluation of,and in the optimisation of waste managementstrategies to meet set objectives.

4 PROCESS SIMULATION: MASS BALANCEPROCESS MODELS

Once the feedstock to any waste management processhas been characterised (by direct measurement orthrough estimations according to the schema presentedabove) simulation techniques can then be applied topredict how this feedstock will actually respond to theprocess. Process, in this context, is defined in its

49

broadest sense and may be anything from a simpleselective collection (source segregation scheme) to afully mechanised central waste sorting plant. Themethodology for the simulation remains the samethroughout.

The technical operation of each unit process,within the overall treatment, can be describedmathematically in terms of a (mass balance) processmodel. Two main classes of model can be defined:(i) a separation model, where waste is separated intotwo or more discrete streams according to physicalcharacteristics (eg. by particle size, density ormagnetic susceptibility) and (ii) a transformationmodel where the waste undergoes physical orchemical change (eg. size reduction, pelletising,incineration etc).

Mass balance models for common mechanised

sortingdevices have been reported in the literature formore than a decade. These published models, whichoften have their origins in mineral processingtechnology, range from being purely theoretical tobeing purely empirical in nature. There is, however,little published data on just how well these modelsactually fit experimental observations. Someevaluation of the quality of the models has recentlybeen undertaken by Warren Spring Laboratory, usingboth pilot scale data and data collected fromcommercial operations for reference. Although theresults from the study can not be published forcommercial reasons, the main conclusion can bedisclosed: 'most models, as they stand, do notprovide an adequate description of the process butthey can be made to do so simply through empiricaltuning.' This finding closely parallels the experiencesof the mineral processing industry, where, arguably,the application of the simulation method has reacheda more advanced stage and has several documentedcommercial applications (eg. 2,3). The'phenomenological' modelling method that has beensuccessfully applied to mineral processing plant, isoutlined below. Reference is made to waste treatment

applications.

4.1 Separation models

In sorting operations (eg. trammelling, magneticseparation, air classification etc.), the particle sizeand/or other physical characteristics of the particlewill be important in determining that particle'sresponse to sorting. The additional physicalcharacteristics will, to a first approximation beimplicit properties of the material type itself and can

50

be regarded as constant within any material type.Paper and card, for example, have relatively lowspecific gravities whereas ferrous metal has a highspecific gravity, the settling rates of plastic film areslow due to particleshape effects, ferrous metals havehigh magnetic susceptibilities etc. The relativeimportance of each of these physical characteristics tothe separation is a function of the separating deviceused. For each separating device, separationperformance can be described in terms of thefractionation of each size/ material class, from theprocess feed <j) to each device output (p).Mathematically this can be expressed as:

Ajk ~~ft) • ^ijk (1)

where the transfer coefficient, T, is a function of thematerial category, the particle size and of how theseparating device is operated. For each device, thefollowing set of relationships can be formulated todescribe the function T.

T9t=fa.(!JJ3...,OV'lt9P) (2)

P = fh.(0VV) (3)

The main model relationship (equation 2) comprisesone or more equations formulated in terms ofcontinuous functions of each size, i, and of eachphysical propertyjB. Each material class is assigned adiscrete value for each property variable, j. The mainmodel relationship also takes into account anyoperatingdependencies (OV) which can be quantifiedanalytically. The effect of many operating variables,however, are very difficult to delineate and candepend, for example, on the exact fine-scale make-upof the waste. It would prove impossible to quantifythese dependencies to high precision, although theunderlying trends can usually be identified withreasonable clarity. These trend equations can beexpressed as an auxiliary model relationship (equation3). The link between the main and auxiliary modelequations is a set of adjustable model parameters, P.

The parameter set P provides the essential meansfor empirically tuning the model to match observation.Establishing the parameter set P provides the devicemodel calibration. Calibration can be effected

mathematically by regressing the model equations ontomeasured data sets, where available, to provide theclosest (least squares) fit. Reference parameter setscan then be stored in a reference library to be drawnon when measured data is not available.

Once the calibration is established, equations 2 and3 can be applied in a predictive sense, to estimateseparation performance under different operatingregimes. This is important, when it comes tooptimising the process in order to meet specificprocessing objectives. This aspect will be developedfurther in section 6.

4.2 Size reduction models

Size reduction models (eg. for knife mills, hammermills etc.), are based, in the first instance, on thebreakage of homogeneous material. The models treatthe size reduction process either as a single event oras a series ofbreakage events. In each breakage event,a fraction ofthe material is regarded as being selectedfor breakage and then broken into a distribution ofsmaller particles; the rest of the material remainingunbroken. The mathematical formulation of thisprocess is usually expressed in a matrix form:

p = B.S/+ (I-S)./ (4)

Many functions have been proposed to fix theindividual values in the selection for breakage matrixS and in the breakage matrix B. An example of thetype of function which is now being successfullyapplied to refuse size reduction is illustrated below.The quoted example refers to a function originallyproposed some forty five years ago by Epstein (4).

&.=, \ Aj.Aj.+1 /

a/2 (5)

The adjustable parameters in the equation are St (thevalue of the selection function for the top size class)and the index, a. These parameters will take differentvalues for each material category and, like theseparation model parameters, must be empiricallydetermined.

Comminution of composite materials often leads tototal or partial liberation of individual materialcomponents. Liberation modelling, which has receivedmuch attention in mineral processing, has proved verydifficult to model accurately. A simple approach,which appears to work reasonably, was developed atWarren Spring Laboratory (5) tor gravityconcentration circuits. It seems reasonable that similarmodels could be developed for operations in wasteprocessing, although these models would probablyneed to be material specific (eg. targeted to a batterybreaking plant, or to scrap wire recovery and so on).

4.3 Waste collection models

Mass balance models of waste collection schemes will

provide a link, where necessary, between the wastearisings data and the waste processing models. Thewaste collection models will take exactly the sameformat as the mechanised sorting models and, as suchcan be fully integrated with them in the processsimulator. The model functions are still adequatelydescribed by equations 1 to 3 above, with T nowrepresenting the diversion rate of a particular wastecomponent to the collection scheme. Contamination ofthe collectibles by other material (eg. tin-plate in analuminium can bank) is represented by non-zero Tvalues for the contaminants. The operating variables,rather than being machine settings, will relate more tothe geodemographic factors associated with thecatchment area.

Prototype (probability) models can be built fromanalysis of the monitored performance data of thevarious collection schemes already in operation.

It must be borne in mind, when applying collectionmodels, that the input (waste generation) data mightitself be a function of the collection scheme.Corrections for this must be applied to the raw datarather than to the model. The modelling of thesecorrections has already been discussed in section 2above.

5 PROCESS SIMULATION: OTHER MODELS

Chemical analysis data can be overlaid onto the massbalance models to enable predictions of the chemicalcomposition, or energyvalue, of each separated wastecomponent. Much of the chemical information that isneeded will be available from the National WasteAnalysis Database, though not always at the level ofdetail required for the simulation (ie. as separateanalyses for each individual size/ material category).Full analyses, where available, will however providebase data relating to the typical chemical partitionswithin the waste. It may then be possible to applythese data more generally and to upgrade partialanalyses through the establishment of suitable scalingfactors.

In the simulation of sorting processes, a good firstapproximation towards solving the chemical balanceis the assumption that the chemical composition ofeach size/ material fraction ofthe sorted product is thesame as that of the equivalent fraction in the processfeed. This implies that each sub-fraction is fairly

51

•)UV itW.1V. .Tif '<:

T>m.v<;A!.I.:0;,-u.;;,.

' •?..';

•;• ;.v;Mv-- i.^; A'.O•'

.5

:?r':.)

-II

homogeneous in content and that separation is notbiased significantly by material variations within thatfraction. The accuracy of the chemical simulation willthus improve when a finer division of materialcategories is used (ie. the categories become morehomogeneous). In this respect a 33 categoryrepresentation should furnish more accuratecomposition predictions than would be achieved byusing a coarser resolution of 11 categories. Thismethod of prediction would, however, be totallyinappropriate for some specialised sorting processes(eg. cadmium partitioning in the selective collection ofNi-Cad batteries), and an alternative approach wouldneed to be found in these instances.

There are also many difficulties in predictingchemical partitioning in size reduction processes. Aswith the mass balance models, the inherent problemsare linked to the possible heterogeneity of thecomminuted fragments. This is .clearly an area forfurther focused fundamental research.

Energy models can also be built for each of theunit processes. In the simulator these energies arecumulated to provide the total power consumptiondata. Energy models and raw consumption data areavailable from the published literature and fromequipment manufacturers. It must be borne in mindhere, that in assessing total energy consumption,material transport requirements should also beincluded within the overall process. This maynecessitate the construction of additional sub-models,eg. for conveyor belts.

Economic models may also be applied. Thesemodels essentially cumulate capital, installation,finance, labour and maintenance costs of processoperations or refer to the revenue and disposal costsof the processed products.

6 PROCESS SIMULATION

Process simulation is the computer representation ofwhat happens, or might happen, in a real process. Inthe current context, process simulation provides adescription of waste treatment operations involvingone or more coupled unit treatment processes (orprocess flowsheet). The flowsheet comprises anetwork, or line, of nodes with connecting flowstreams. Each node denotes a unit process,represented within the simulator by a mass-balanceprocess model. The mathematical solution of theflowsheet adopts techniques originally developed inthe chemical engineering industry. A review of these

52

techniques is given by Westerberg et al. (6). Solutionof the flowsheet involves partitioning the circuit intoits constituent loops and iteratively solving each loopand each of the models contained therein. As thewaste management models are generally non-linear informat, and cannot easily be linearised, a serialsolution method is preferred.

The Warren Spring Laboratory simulator GSIM(5), developed originally for mineral processingapplications, contains two important additionalfeatures: (i) flow constraints can be set and (ii) thecircuit operation can be tuned to optimise overallprocess performance against a given objectivecriterion.

A large number of variables can affect circuitperformance. To establish the optimum combinationof these variables requires the application of anumerical search algorithm. In GSIM, a pattern searchmethod has been chosen (4), although a derivative-based method could equally suffice. The objectivecriterion against which the process performance isjudged will vary according to application. In wasteprocessing, this objective function might be tomaximise energy recovery, to maximise recovery ofa recyclable component or of the total recyclables andso on. Constraints might also be set, for example acost ceiling, or an emission limit or a 25% recyclingtarget. Care must be exercised when setting theseobjectives depending on the quality of the modelsthemselves. Optimising on technical objectives shouldgenerally produce more robust results than wouldoptimising against economic objectives.

7 STATE OF THE ART

The concept presented in this paper is ambitious. Aprototype (validated) integrated system is, however,not that far from realisation. Much of the modellingand database technology has already been developed,some of it as much as forty five years ago. Also,enough data now exists, both with regard to wastecharacterisation and with regard to process orcollection performances, to make applicationindustrially viable. The innovation lies in assemblingthe technologies into, an integrated whole andvalidating the methods against documentedmeasurement.

There are still, of course, many gaps in the dataand in our overall understanding. As knowledgegrows, these gaps can be progressively filled. Forexample, civic amenity waste and commercial and

^r-X^tf :M

industrial wastes are not categorised to the sameextent as domestic wastes. A General Industrial WasteAnalysisProgramme, sponsored by DTI, is, however,now under way and should furnish some of thisinformation. Further fundamental work needs to becarried out on model development, to introduce awider range of models and to refine the currentmodels. Particular attention needs to be given tobroadening the validity of chemical partition models.

REFERENCES

(1) ANON. National household waste analysisproject. Department of the Environment:Project profile 009, 1992.

(2) LEWIS, K.A, WELLS, A, TUCKER, P,Computer simulation at Wheal Jane,Cornwall. Proceedings ofMineral Processingin the United Kingdom, Leeds, 1989,95-105.

(3) TUCKER, P, LEWIS, K.A., WOOD, P,Computer optimisation of a shaking table.

Minerals Engineering, 1991, 4,355-367.(4) EPSTEIN, B, The mathematical description of

certain breakage mechanisms leading to thelogarithmico-normal distribution. Journal ofthe Franklin Institute, 1947.224

(5) MACKIE, I, TUCKER, P, Simulation andOptimisation ofgravity concentration circuits.Proceedings of International Symposium onthe Mathematical Modelling of MetalProcessing Operations, Palm

Springs,\m, 147-160, SME: Warrendale,PA.(6) WESTERBERG, A.W., et al., Process

flowsheeting, 1979,Cambridge UniversityPress:Cambridge.

ACKNOWLEDGEMENT

The National Domestic Waste Analysis Programmehas been funded by the UK Department of theEnvironment and INCPEN.

© Crown Copyright 1993

53

t*••••.». . ir.- i. %v A-v ....ir.'./ .' .c .

I..' T:

Hi

:.J;-. »?•.•; "3 !

AG

54

1

SPECIFIC

WASTE

ANALYSES

i , i

^..sw.vw.vaw.v.'av.-.was-.wwav

ENERGY

SUB

MODELS

wmx::xx;xsxzA

NATIONAL

WASTE

STATISTICS

DATABASE

STATISTICAL

MODEL OF

WASTE

MASS BALANCE MODEL

1

! (SELECTIVE)| COLLECTION

MODELSi

upl

up2 up3

up3 up5

up4 up6 • up7

FLOWSHEET OF

UNIT PROCESS MODELS

PRODUCT SPECULATION

SOCIO

ECONOMIC

PROFILE

SEASONAL &

COLLECTION

FACTORS

1 '

Fig 1 Schematic representation of waste management model

-