Sei sulla pagina 1di 118

Modeling of Material Recovery Facility Performance with

Applications for Life Cycle Assessment


by
Anne G. Raymond
B.A., Chemistry
Dartmouth College, 2006

Submitted to the Institute for Data, Systems, and Society and the Department of Mechanical
Engineering in Partial Fulfillment of the Requirements for the Degrees of

Master of Science in Technology and Policy and Mechanical Engineering

at the

Massachusetts Institute of Technology

June 2017

@ 2017 Massachusetts Institute of Technology. All rights reserved.

Author: Signature redacted__


Institute for Data, Systems, and Society
Department of Mechanical Engineering
May 24, 2017

Certified by: Signature redacted


( UJ Timothy Gutowski
Professor of Mechanical Engineering
Srrd a Thdis Supervisor
Accepted by:
Munther Dahleh
Professor of Electrical Engineering and Computer Science
Director, Institute for Data, Systems, and Society

Accepted by: Signature redacted


MASSACHUSETTS INSTITUT E 2kbhan Abeyaratne
OF TECHNOLOGY Professor of Mechanical Engineering
Chairman, Committee on Graduate Students
JUN 052017
LIBRARIES
ARCHIVES
Modeling of Material Recovery Facility Performance with
Applications for Life Cycle Assessment
by
Anne G. Raymond

Submitted to the Institute for Data, Systems, and Society and the Department of Mechanical
Engineering on May 24, 2017, in Partial Fulfillment of the Requirements for the Degrees of
Master of Science in Technology and Policy and Mechanical Engineering

Abstract
Single stream recycling programs improve participation rates and divert more end-of-
life products from landfill to recycling. The resulting mixture of waste must be sorted, however,
before individual materials can be reprocessed for recycling. Material recovery facilities
(MRFs) use a series of mechanical and manual sorting steps to separate the mixed stream of
materials. The ability of MRFs to correctly sort materials is therefore a critical component of
recycling systems.
In this thesis, we apply a network flow model to evaluate the performance of sorting
facility scenarios. We use experimental sorting efficiencies obtained through waste sampling at
a functioning MRF, and explore how those efficiencies change under different operating
conditions. We further apply the model to find the best configuration of a series of sorting units,
optimizing for both profit and overall sorting efficiency.
Recycling is promoted by the EPA and other organizations as a means of reducing the
impact of material use. The benefits of recycling relative to landfill and incineration are
determined using life cycle analysis (LCA), which takes into account material recovery rates.
We apply modeled recovery rates for selected materials to a recycling LCA, and show that they
directly impact the magnitude of estimated benefits.

Thesis Supervisor: Timothy Gutowski


Title: Professor of Mechanical Engineering

3
4
Acknowledgments
An enduring lesson I will take away from the past three years is that stretching one's
brain is best done in good company. Fortunately, my time at MIT has included a cast of mentors,
friends, and family who have challenged me to tackle tough questions, trusted my decisions
when I still had doubts, and patiently supported me through the ups and downs of graduate
school.
I am grateful to my advisor, Professor Timothy Gutowski, for consistently making me
think, and for giving me the opportunity to stay for a third year. I would also like to thank
Professor Stephen Graves for his guidance, Mapi for her consistent high spirits and general
brilliance, and Karine for her calm presence and willingness to explain things to me. To the rest
of EBM, thanks for making our office a friendly, welcoming place, I am fortunate to have been
based in 35-005.
That I found my way to the Technology and Policy Program is one of my greatest
fortunes. To Barb, Frank, Ed, Noelle, and everyone at TPP, thank you for introducing me to
new ways of thinking about the world and my role in it.
Cycling team, you gave me some of my most treasured friends and MIT memories, and
a vital source of perspective when school was hard.
A few words cannot adequately thank my parents, who have provided unconditional
love and support throughout my life and from whom I inherited my enthusiasm for both science
and writing. Peter, thanks for being the best little brother a person could ask for. And finally,
Daniel, thank you for always listening, and for your unwavering confidence in me.

5
6
Contents
1 Introduction...................................................................................................................................... 13
1.1 M unicipal Solid W aste in the United States................................................................. 14
1.2 M SW M anagem ent............................................................................................................. 15
1.3 Introduction to Recycling .............................................................................................. 17
1.3.1 Recycling Processes and Definitions ........................................................................... 17
1.3.2 Recycling Collection and Sorting System s ................................................................... 19
1.4 M otivation: The Im pact of Recycling ............................................................................ 21
1.4.1 Potential Energy and Em issions Savings from Recycling .......................................... 22
1.4.2 Role of Material Recovery Facilities in Calculating Benefits from Recycling ........... 23
1.5 O utline for the thesis ....................................................................................................... 26

2 M odeling M aterial Recovery Facilities ....................................................................................... 27


2.1 M etrics of M RF Perform ance ....................................................................................... 28
2.2 Sorting Equipm ent in an M RF..................................................................................... 31
2.2.1 M RF Term inology and Sorting Equipm ent ................................................................ 31
2.2.2 Principles of Separation for Select Equipm ent ............................................................ 32
2.3 M RF M odel and Unit Process Sorting Efficiency ....................................................... 35
2.3.1 Experimental Determination of Sorting Efficiencies for MRF Model ........................ 39
2.4 M RF Input Com position................................................................................................ 43
2.4.1 Characterization of Input M aterial .............................................................................. 44
2.4.2 Composition of Single Stream Recycling in the United States ................................... 46
2.4.3 Variation in W aste Com position................................................................................... 47
2.5 Variation in Sorting Efficiency .................................................................................... 49
2.5.1 Design of Experiments to Test Sorting Efficiency Variation ..................................... 51

2.5.2 M agnet Test Results..................................................................................................... 53


2.5.3 Eddy Current Test Results ........................................................................................... 55

2.5.4 Near Infrared Detector Test Results............................................................................ 56

3 O ptim ization of M RF Design.......................................................................................................... 63


3.1 Evaluation of Single Stream M RFs .............................................................................. 63
3.1.1 M odeled M RF Configurations .................................................................................... 64

3.1.2 Single Stream M RF M odel Scenarios.......................................................................... 66


3.1.3 M odeled Recovery and Grade for Scenarios .............................................................. 69

7
3.2 M RF Performance M etrics............................................................................................ 73
3.2.1 Calculation of Profit..................................................................................................... 73
3.2.2 Calculation of Overall Sorting Efficiency .................................................................. 78
3.2.3 Perform ance of M RF Scenarios ................................................................................... 78
3.2 O ptim ization of M RF Perform ance............................................................................... 80
3.2.4 M odeled Best M RF Design.......................................................................................... 82
3.2.5 M odeled Best M RF with Recirculation ....................................................................... 83
3.2.6 Perform ance of Optimized Configurations ................................................................... 84
3.3 Design Principles for M RFs............................................................................................ 86

4 Application of MRF Model Results to Recycling Life Cycle Assessment............................... 89


4.1 Life Cycle Assessm ent for Recycling ............................................................................ 89
4.1.1 Uncertainty in W aste M anagem ent LCA ..................................................................... 92
4.1.2 Assum ptions in Recycling LCA .................................................................................. 93
4.1.3 LCA M odels..................................................................................................................... 97
4.2 Material Recovery and GHG Emission Factors in WARM........................................ 98
4.2.1 M aterial Loss in Sorting and M anufacture .................................................................. 98
4.2.2 Application of M odel Results to GHG Em ission Factors.............................................. 102
4.2.3 Lim itations to Using the M RF M odel in Life Cycle Analysis....................................... 103
4.3 Im plications for Recycling Policy ................................................................................... 104
4.3.1 M RF M odel Users.......................................................................................................... 105

5 Conclusion ...................................................................................................................................... 107


5.1 Sorting Efficiencies and Variation ................................................................................. 107
5.2 Optim ization of M RF Design .......................................................................................... 107
5.2.1 Pricing and Design for Profit ......................................................................................... 108
5.3 Recovery in LCA .............................................................................................................. 108
5.3.1 Implications for Policy ................................................................................................... 108
5.4 Future W ork ..................................................................................................................... 109
5.4.1 Autom ated W aste Com position M easurement .............................................................. 109
5.4.2 M odel of Price Data for Varied Grades of M aterial ...................................................... 110

Appendix ............................................................................................................................................ 111

References........................................................................................................................................... 115

8
List of Figures
Figure 1-1: Material composition of MSW in the United States in 2013.............................................. 15
Figure 1-2: EPA Waste Management Hierarchy ................................................................................ 16
Figure 1-3: M aterial life cycle flow s .................................................................................................. 18
Figure 1-4: Collection schemes for municipal solid waste recycling ............................................... 20
Figure 2-1: Sorting step w ith tw o output units .................................................................................. 28
Figure 2-2: A simple MRF with three binary sorting units .............................................................. 30
Figure 2-3: O ver belt m agnetic separator ......................................................................................... 33
Figure 2-4: Eddy C urrent Separator................................................................................................... 33
Figure 2-5: N ear Infrared (N IR) D etector.......................................................................................... 34
F igure 2-6 : B allistic separator................................................................................................................ 35
Figure 2-7: Sorting efficiency of paper in a two-step sorting process............................................... 38
Figure 2-8: Light packaging recovery section (LPRS) of Spanish MRF.......................................... 40
Figure 2-9: Quarters procedure for sampling heterogeneous waste ................................................ 45
Figure 2-10: Variations in composition of input waste at a single stream MRF in York, PA........... 48
Figure 2-11: Variations in the recyclable component of mixed waste input to a section of the Spanish
M RF over five sets of sam ple data............................................................................................ 49
Figure 2-12: Recovery and grade of ferrous metals under varied target material concentration and
m agn et h eig ht ................................................................................................................................ 53
Figure 2-13: Cylindrical and pressed soda cans ................................................................................ 54
Figure 2-14: Separation efficiency for ferrous and non-ferrous material versus concentration of ferrous
in sam p le ....................................................................................................................................... 55
Figure 2-15: Recovery and grade of aluminum output under varied aluminum concentration and
rotation speed of the eddy current m agnet ................................................................................ 56
Figure 2-16: Results of input sample variation NIR test: recovery of PET, grade of target output, and
fraction of PET in input sam ple ................................................................................................ 58
Figure 2-17: Composition of contamination in PET outputs for five different input samples.......... 59
Figure 2-18: Recovery and grade of PET output from test NIR using sample A .............................. 61
Figure 3-1: Single stream MRF Configuration A .............................................................................. 65
Figure 3-2: Single stream MRF Configuration B ............................................................................. 66
Figure 3-3: Input composition for model scenarios 1-4 (EPA) and 5 (Pennsylvania)...................... 67
Figure 3-4: Composition and value of materials in single stream MRF input................................... 75

9
Figure 3-5: Modeled selling price of materials as a function of grade with a purity threshold of 0.75 77
Figure 3-6: Unordered units for use in the optimization function of the MRF model....................... 81
Figure 3-7: Optimized configurations for Scenario 3 sorting equipment and input.......................... 83
Figure 3-8: Optimized MRF designs for MRF in Scenario 3 with the addition of an NIR unit (U 10)
w ith no specified target m aterial. ............................................................................................... 84
Figure 4-1: Displaced production of a recycled product .................................................................. 95

10
List of Tables
Table 1-1: The embodied energy and CO 2 footprint of recyclable materials ................................... 23
Table 2-1: Estimated target material sorting efficiencies for single-stream MRF............................. 37
Table 2-2: Composition of the input to the Light Packaging Recovery Section magnet for five different
sam p le s .......................................................................................................................................... 41
Table 2-3: Experimental separation efficiencies for LPRS sorting equipment ................................. 42
Table 2-4: Materials in generated and recycled in the United States in 2005, 2010, and 2014.......... 43
Table 2-5: Composition of residential single stream recycling ....................................................... 47
Table 2-6: Magnet Test 1.1 (normal ferrous input concentration and magnet height)...................... 52
Table 2-7: Sorting efficiencies for ferrous and non-ferrous materials in Magnet Test 1.1 ............... 52
Table 2-8: Composition of five input samples for NIR Test 1 ......................................................... 57
Table 2-9: Sorting efficiencies for materials in target output stream of NIR unit selecting PET.......... 60
Table 3-1: Scenario 1 target stream separation efficiencies .............................................................. 67
Table 3-2: Scenarios 2 - 5 separation efficiencies............................................................................ 68
Table 3-3: Ballistic sorting efficiencies for Scenarios 3 - 5.............................................................. 69
Table 3-4: Modeled recovery and grade for MRF scenarios 1, 2, 3 and 4 ........................................ 70
Table 3-5: Modeled recovery and grade of materials for Scenario 5 ................................................ 72
Table 3-6: Selling price of recovered materials in dollars per metric ton .......................................... 74
Table 3-7: Overall sorting efficiency, revenues, and costs for Scenarios 1-5 ................................... 79
Table 3-8: Sorting, output, and input units for model scenarios........................................................ 80
Table 3-9: Performance metrics and output stream data for modeled best MRF designs in
O ptim izations A - D . .................................................................................................................... 85
Table 4-1: Emissions from virgin and recycled aluminum and steel can production......................... 91
Table 4-2: Global average recycled content of m etals........................................................................ 94

Table 4-3: EPA WARM sorting, manufacturing, and overall retention rates for materials in MSW ... 99
Table 4-4: Comparison of recovery rates for selected materials used in WARM with modeled and
m easured recovery rates and target sorting efficiencies ............................................................. 100

11
12
Chapter 1

Introduction
The useful lives of materials extend beyond the lives of the products they make up. In
countries with developed waste management systems, discarded materials may be sent to
landfill, biodegrade in a composting facility, be reused by the consumer or someone else,
contribute to electricity generation in a waste-to-energy facility, or be recycled and incorporated
into a new product. Each of these pathways has associated energy use and greenhouse gas
emissions. As concern about resource scarcity and material sustainability grows, it is
increasingly important to be able to quantify the environmental impact of what happens to
consumer products when they are thrown away so that governments, companies, and
individuals can make informed decisions about how to manage waste.
Recycling is the waste management option that seems to offer a way for society to
reduce the environmental impact of consumer products without reducing consumption. Free
curbside collection of mixed recyclables in many communities in the United States has made
recycling easy for consumers, and recycling is promoted as a way for people to be responsible
about their material use. Though simple in concept, however, recycling is a complex process
that involves many different steps and design choices, all of which determine how much energy
is saved when, for example, a consumer puts a plastic bottle into a recycling bin rather than the
trash.
This thesis addresses two key questions related to recycling. The first is how effectively
materials that consumers toss together into one bin can be separated by mechanical sorting
facilities. This is important because while it is easier for people to recycle when they do not
have to put time and effort into sorting, waste materials eventually have be sorted before they
are incorporated into new products. The trend in the United States and Europe is to carry out
this sorting using machines, but there is little data on their performance. The second question
is whether a better understanding of how mechanical sorting works changes estimates of the
overall environmental impact of recycling. Calculating the energy and greenhouse gas

13
emissions associated with any of the end-of-life options for materials requires a series of
assumptions. For recycling, how well materials are sorted is one of those assumptions.
This chapter introduces the state of waste generation and recycling in the United States,
and the options for end-of-life materials management. The steps in the recycling process are
outlined as well as the principles that underlie the environmental benefits of recycling.

1.1 Municipal Solid Waste in the United States

Municipal Solid Waste (MSW) consists of household items that people throw away after
use, including packaging, yard waste, food, and electronics. MSW includes waste from homes,
schools, hospitals, and other commercial locations, but excludes hazardous waste, and that from
industry and construction. In 2014, the United States generated 234.6 million metric tons of
MSW, or about 2 kilograms of waste per person per day (U.S. EPA 2016a). We note that tons
in this thesis always refers to metric tons.
The composition of those many millions of tons of waste provides an interesting view
into the types of materials and products consumed in the U.S. In 2014, about 30% of MSW
consisted of packaging, and another 20% was non-durable goods, or products designed to last
less than three years such as newspapers, clothing, and shoes. Figure 1-1 shows that almost half
of household waste consists of commonly-recycled materials (paper, glass, metals, and
plastics), and a second large fraction, organic food and yard waste, is compostable. According
to the United States Environmental Protection Agency (EPA), however, in 2014 only 26% of
MSW in the United States was recycled. In the same year, 8.9% of waste was composted, 12.8%
was incinerated with energy recovery, and the remaining 52.6%, was landfilled (U.S. EPA
2016a). Given the amount of recyclable material in the waste stream, these data suggest that
current recycling rates are well below their possible limit.

14
Moir.

Other
3%

Glass
Metals-4%
9%
Wood 9
6% Rubber,
leather,
&

Textiles
10%

Figure 1-1: Material composition of MSW in the United States in 2013 (U.S. EPA 2016a)

Not all of the products in the waste stream that are made of recyclable materials are
actually able to be recovered and recycled, so the generation of materials is an upper bound for
the actual recycling rate. There are several factors that impact whether or not a material is
recycled. One is consumer behavior and available municipal services. If a community or
institution does not provide recycling services, or if people simply do not make use of them,
then plastic bottles, aluminum cans, newspapers, and other recyclable materials will go to
landfill. Even when products and materials are collected for recycling, they may be hard to
separate from other types of material, or may be unsuitable for manufacture into new products
due to moisture, presence of food waste, or dyes. Calculating the fraction of materials collected
for recycling that are recovered by sorting is the focus of much of the rest of this paper. As a
starting point, however, the generation of waste provides an estimate for the order of magnitude
of unrealized MSW recycling.

1.2 MSW Management

Concerns about human health and safety have pushed the development of waste
management systems. Landfills, once simply dumping areas for trash, are now regulated by the
EPA through the Resources, Conservation, and Recovery Act (RCRA), and employ innovations

15
such as impermeable linings, leachate monitoring, and methane capture to reduce their local
impact on air and water quality (U.S. EPA 2017). Although modern landfills may be carefully
managed to reduce local impacts on water and air quality, they are not an efficient use of
materials. As Figure 1-1 showed, most of MSW in the United Sates is made up of materials
that, even at end-of-life, have some use or value. For example, food waste can be composted,
and the resulting compost used to fertilize farmland. Packaging such as aluminum cans or
plastic bottles can be recycled into new cans and bottles, or into other products including, for
plastic, fleece clothing, car bumpers, and children's toys. Products such as mobile phones and
other electronics can be refurbished or valuable components salvaged for reuse.
When materials enter the waste stream, there are three primary alternatives to sending
them to landfill: reuse, recycling, and combustion or incineration for energy recovery. The order
of these options in the EPA's Waste Management Hierarchy in Figure 1-2 reflects decreasing
preference based on increasing environmental burden. The EPA calculates the environmental
impact of each scenario using the Waste Reduction Model (WARM), a tool which quantifies
the energy and carbon emissions associated with the steps from the point of waste generation
to final end use or disposal. The calculation of the benefits of recycling using WARM are
addressed in more detail in Chapter 4.

1.
Source Reduction and Reuse

Recycling and Composting

Combustion with
Energy Recovery

Landfill

Figure 1-2: EPA Waste Management Hierarchy (U.S. EPA 2017b)

While the focus of this work is on recycling, the costs and benefits of recycling are best
understood in comparison with other waste management options. The top choices in the waste

16
hierarchy, reduce and reuse, have obvious benefits from lowering material demand for products
and the amount of waste discarded. Combustion with energy recovery captures the chemical
energy of waste through incineration, and uses the heat from combustion to generate electricity.
The relative preference for combustion with energy recovery over landfill is based on the
substitution of electricity generated by waste combustion for existing power generation from
other sources. The benefits of incineration depend significantly on the type of materials, and
therefore the energy content, in the waste being incinerated, and the carbon emissions
associated with existing power generation. For example, in areas where electricity is generated
primarily by coal-fired power plants, replacing some generation capacity with a waste-to-
energy plant can reduce overall carbon emissions. If the local electricity mix is largely
renewable, however, then combustion with energy recovery would increase the carbon intensity
of electricity, and landfill may be the better alternative for the waste.
Composting may be carried out at the household level, or on an industrial scale. One of
the primacy challenges of large-scale composting is ensuring the quality of incoming waste, as
even small amounts of hazardous or non-biodegradable materials can contaminate a large
quantity of otherwise useful organic matter. In most cases, high-quality compost requires
source-separated organic matter i.e., food or yard waste that is collected separately for
composting, and never mixed in with other types of waste. Effective composting systems may
complement recycling, as organic matter is a contaminant in recycling streams. If food waste is
collected separately from packaging, containers, paper, and other recyclables, it improves the
outputs of both processes.

1.3 Introduction to Recycling

This section defines some useful terms, describes the systems used to collect recyclable
materials, and introduces the facilities where mixed recycling streams are separated.

1.3.1 Recycling Processes and Definitions

Recycling itself is a broad term that the EPA defines as "recovering and reprocessing
usable products that might otherwise become waste" (U.S. EPA 2016c). Recycling rates and

17
related terms are defined inconsistently in the literature. We adapt the terms used for metal
recycling by Graedel (2011) to the life cycle of materials in MSW more generally. This
approach overlooks the details of material-specific life cycles, but is useful as a simple model.
The material life cycle illustrated in Figure 1-3 provides the basis for defining end-of-life
metrics for recycling. Note that sorting and reprocessing are often combined into one recycling
step, however as we are interested specifically in material sorting, we have divided them into
two distinct processes.

Product A
............................................................................................................................................................................

Manf acture Collection Row


Production Fabrication

(h)

Sorting

.
Reprocessing s

. . I.. .. .. . . . ................................ .........

*Life cyle of

*Product B

a) primary material input g) EOL material (recyclate) retained for


b) refined material closed-loop recycling
c) intermediate product h) scrap from manufacturing
d) end-of-life (EOL) product i) new and used scrap used in fabrication
e) EOL material collected for recycling J) new and used scrap used in production
(Collected) k) EOL material retained for open-loop
f) EOL material recovered by the sorting recycling
process (Recovered)

Figure 1-3: Material life cycle flows (adapted from Graedel 2011, 16). Material losses are indicated by
dashed arrows, and the system boundaries for each product are the dashed borders.

18
The metrics for end-of-life material recycling are calculated using the mass flows in Figure 1-3
as follows:
Collection rate: C = eld (1-1)
Sorting recovery rate: R = f/e (1-2)
Reprocessing retention rate: P = g/f (1-3)
End-of-life recycling rate (closed loop): EOL-RR = g/d (1-4)
End-of-life recycling rate (open loop): EOL-RRo = k/d (1-5)
Recycled content: RC = (i + j)/(a + i + j) (1-6)

Materials that are not diverted in the collection, sorting, and reprocessing steps are
shown as sent to landfill, however they could also be incinerated or composted, depending on
the material composition. As recovery is a key metric of the sorting process itself, but also a
general term, some additional clarification is useful to avoid confusion. Material stream g is
defined by Graedel as the recyclates, or those materials able to be reused after reprocessing. We
will use recyclates to distinguish stream g, the materials retained throughout collection, sorting,
and reprocessing, from streamf, the materials recovered by only the sorting process.

1.3.2 Recycling Collection and Sorting Systems

One of the challenges for recycling system managers is how to collect and separate
recyclable end-of-life materials from other waste. Communities generally use one of the four
strategies illustrated in Figure 1-4. The first is source separation, where consumers separate
their recyclables by material type. Source separation usually requires individuals to transport
their recyclables to a local transfer station or recycling center where they can place items into
the appropriate collection bins. Container deposit laws, or bottle bills, promote source
separation for the specified containers. Source-separated materials are the simplest to recycle,
as they can be sent directly to reprocessing without an additional separation step, but the system
requires effort on the part of consumers. If a community provides weekly curbside trash
collection, as many do, there may be little incentive for consumers to take the time to drive to
a recycling collection center.

19
Source Separated Recycling Double Stream Recycling Single Stream Recycling Mixed Waste
Plastic _______________
Newsprint Cardboard (PET)
rPaper and Cardboard All Household Waste
Aluminum Patc_
metals (HDPE) _____________

'61' Office i Plasticiic Plastic, Metal, Glass Cardboard, Plastic, Mixture of recyclables,
i Metal, Glass and other trash
Paper (others) M Gcompost,
sorted at mixed waste MRF
Littleor no post-consumer sorting required Commingled recycling streams sorted at an MRF

Figure 1-4: Collection schemes for municipal solid waste recycling

The other three strategies shift the burden of sorting recyclables downstream from the
consumer. MSW that is collected without any source separation, i.e., with all compost,
recyclables, and trash together in one bin, is called mixed waste. Due to the variety of materials
in mixed waste, effective separation of the different materials is challenging, and this system is
not prevalent in the United States (U.S. EPA 2015). The more common approach is to collect
mixed recyclables separately from other waste, either in one or two streams. In dual-stream
recycling, consumers separate recyclables into two bins for curbside collection, one container
for paper and cardboard, and a second for other packaging, including plastic bottles and metal
cans. Single-stream or single-sort recycling requires only that consumers throw all recyclables
into a single bin for curbside collection. Single stream recycling is increasingly popular in the
United States (Fitzgerald, Krones, and Themelis 2012). In 2011, almost 70% of the U.S.
population lived in communities with curbside recycling collection programs, of which about
65% were single stream (U.S. EPA 2015).
Waste that is collected in mixed waste systems or single and double-stream recycling
must be separated into its component materials for recycling and composting. A material
recovery facility (MRF) uses a combination of mechanical sorting equipment and personnel to
separate desirable materials from each other, and from other waste. This other waste, called
residue, is sent to landfill. The EPA reports that there were 797 MRFs in the United States in
2014, with an estimated total throughput of 127,320 tons per day. In the same year, there were
80 active waste-to-energy facilities, and 1,908 landfills (U.S. EPA 2016b).

20
1.4 Motivation: The Impact of Recycling

In 2015, the United States generated 6,587 million metric tons of carbon dioxide
equivalent (MMTCO 2 E) emissions. Carbon dioxide equivalents are used to report the total
global warming potential (GWP) from a mix of different greenhouse gases. Each gas has a
different GWP depending on its radiative forcing effects over a given time horizon. The GWP
of CO 2 over 100 years is 1. For the same time horizon, methane has a GWP of 25, and for N 2 0
it is 298. The 2015 emissions represent 2.3% decrease from the previous year, and gross annual
emissions are still well below their peak in 2007 (U.S. EPA 2017a). The country's nationally
determined contributions to the Paris Agreement on climate change, however, are to achieve at
least a 26% reduction in GHG emissions over 2005 levels by 2025 (International Energy
Agency 2016). Reaching this target means cutting annual GHG emissions by more than 1,000
MMTCO 2E over the next 8 years.
Industry, including material production, accounts for about 21% of U.S. GHG
emissions. The largest contributors to total emissions are electricity generation (29%), and
transportation (27%) (U.S. EPA 2017a). There is overlap between these sectors, as some
electricity and transportation is related to moving materials around, and one EPA estimate puts
the fraction of 2006 emissions associated with the manufacturing, use, and disposal of products
and materials at 42% (U.S. EPA 2016c). The production and use of materials is therefore a
significant contributor to emissions, and reaching the targets set out in Paris, or any sustainable
emissions future, will require a reduction in material-related energy use.
One way to reduce emissions is to divert more materials to recycling rather than landfill.
The EPA estimates that the impact of recycling 53 million tons of paper, cardboard, glass,
metals, and plastics from MSW in 2014 resulted in a savings of 165.3 MMTCO 2 E, or the
equivalent of taking 35 million passenger cars off the road (U.S. EPA 2016a). This is 14% of
the 1,175 MMTCO 2E reduction needed to bring the country from current emission levels to the
2025 emissions target. The scale of the estimated benefits of recycling is therefore relevant to
national and global reductions in GHG emissions. The most effective way to reduce materials-
related emissions would be to reduce total consumption of products, following the EPA's
reduce and reuse guideline. Under realistic consumption rates, however, these figures show that

21
recycling as an alternative to landfill will be an important component of how the United States
achieves a lower-carbon future.

1.4.1 Potential Energy and Emissions Savings from Recycling

The environmental impacts of different waste management strategies are evaluated


through life cycle assessment (LCA). An LCA accounts for the energy costs of a material or
product over its entire lifecycle, which begins when the raw materials are extracted from their
natural state, and ends with one of the waste management strategies described above. It is well
known that for consumer products requiring significant energy to operate, such as cars or
kitchen appliances, the impact of the use phase dominates the overall life cycle energy and
emissions. Efforts to reduce the emissions from passenger vehicles, for example, focus largely
on fuel efficiency, as over a vehicle's lifetime the total energy associated with gasoline far
exceeds the energy needed to manufacture the car. For packaging and other non-durable goods
that make up most of the recyclable component of MSW, however, the use phase is relatively
short and requires little energy. Instead, it is the materials and manufacturing that dominate the
life cycle emissions (Ashby 2013).
Recycling can lead to an energy and emissions savings over landfill because the energy
needed to produce recycled materials is less than the energy associated with new materials. In
other words, the embodied energy of recycled material is less than for virgin material. The
embodied energy of a material is the energy needed to make 1kg of the material (Ashby 2013).
It is measured in units of energy per unit mass of material, and accounts for energy needed to
extract, refine, and process a material. It is expressed in units that allow the embodied energies
of materials made using different forms of energy to be compared directly, such as megajoules
oil equivalent (MJoe). Data for the embodied energy of the virgin and recycled form of some
common recyclable materials are found in Table 1-1. Due to the assumptions involved in
calculating these values, they are estimates with a standard deviation of 10% at best. For
values to be meaningfully different, they must differ by 20% or more (Ashby 2013). These
values are therefore not appropriate for exact calculations, but provide a basis for comparison
between using virgin and recycled materials to make products. For example, the table shows
that the embodied energy of 1kg of virgin polyethylene terephthalate (PET), the plastic used to
make disposable water bottles, is about twice that of 1kg of recycled PET.

22
Table 1-1: The embodied energy and CO 2 footprint of recyclable materials (Ashby 2013)

Embodied Energy CO 2 Footprint


(MJ,/kg material) (kg C0 2/kg material)

Virgin Recycled Primary Recycling


Material Material Production
Paper and Cardboard 51.5 19.5 1.2 0.8
Glass 10.5 8.2 0.8 0.5
Steel (Ferrous) 26.5 7.3 1.8 0.4
Aluminum 210.0 26.0 12.0 2.1
Copper 58.0 13.5 3.7 0.8
Lead 27.0 7.4 2.0 0.5
Polyethylene terephthalate (PET) 85.0 39.0 3.9 2.4
Polypropylene (PP) 79.5 50.0 3.1 2.4
Polyethylene (PE) 81.0 50.0 2.8 2.9

Studies of waste management LCA conclude that under most conditions, and for most
materials, recycling yields energy savings and lower CO 2 emissions than incineration or landfill
(Bjrklund and Finnveden 2005). This is expected given the differences in embodied energy of
virgin and recycled materials, and confirms that recycling is a key part of sustainable materials
management. The magnitude of the benefits from recycling, however, is less certain. Use of
LCA to evaluate the impact of waste management options requires careful drawing of
boundaries, and assumptions about each step in the life cycle. Variations in these boundaries
and assumptions yield different results. Some of the complexities of recycling LCA will be
explored in the context of the EPA's Waste Reduction Model (WARM) in Chapter 4.

1.4.2 Role of Material Recovery Facilities in Calculating Benefits from Recycling

The calculated benefits of recycling are sensitive to the assumptions in LCA models.
Improving those assumptions is therefore an important step towards more reliable modeling of
the impact of different waste management scenarios, particularly recycling. A greater
understanding of the costs and benefits of waste management, as well as the details of material
sorting and processing, will support local decision-makers in choosing between different
management options for specific communities. This could enhance the efficiency and cost-
effectiveness of waste management systems, and, on a larger scale, improve overall material

23
sustainability. Furthermore, more robust inputs to waste LCA will improve the credibility of
the LCA results, making them more useful to policy makers interested in improving regulations
governing the sustainable management of waste in the United States.
An important step in the life cycle of recyclables collected in a single stream is sorting
at an MRF. This step requires energy to transport materials to the MRF, and to run the facility,
but the more important factor is how well the waste is sorted at the MRF. The performance of
the MRF determines both how much material is recovered from the waste stream, and the
quality of those recovered materials. The fraction of incoming materials that are recovered
determines the ratio between the benefits of recycling, which are based on the recovered
materials, and the original amount of waste. The quality of the recovered materials affects
whether they can substitute for the same type of virgin material in closed-loop recycling, or
they are down-cycled to produce a lower-quality product, yielding fewer benefits.
The energy required to transport waste from curbside collection to MRFs, and to operate
the sorting equipment and facilities, is well characterized, and suggests that single stream
recycling systems are more energy efficient on a per ton basis than double-stream. Primary data
from MRFs in the United States yield an electricity use estimate of 4.7 - 7.8 kWh per metric
ton of input waste for single stream MRFs (Pressley et al. 2015). Changing from double to
single-stream recycling, and therefore using a larger, more complex sorting facility to separate
a greater number of materials, increases the energy required for sorting materials by 19%. This
increase is balanced on a per ton basis, however, by the larger throughput (Fitzgerald, Krones,
and Themelis 2012). There is an additional energy savings from the collection of single-stream
recyclables. Since all recyclables are collected together in one bin, truck capacity can be
maximized, and fewer total truck trips are required than for double-stream (Chester, Martin,
and Sathaye 2008). Increased consumer participation rates with single stream recycling also
yield greater amounts of materials collected for recycling. Materials provide energy benefits
from avoided virgin material use, so a higher fraction of materials diverted for recycling results
in greater benefits (Chester, Martin, and Sathaye 2008; Fitzgerald, Krones, and Themelis 2012).
On a waste management systems level, the energy requirement for electricity to run an
MRF is small compared with fuel costs for collection and transportation (Jaunich et al. 2016).
Even at an MRF, the diesel emissions from on-site vehicles produce far more GHG emissions
than the electricity used to operate the sorting equipment and lighting, and the electricity

24
consumption at an MRF has little impact on overall GHG emissions (Combs 2012). A
parametric analysis of waste composition and equipment sorting efficiencies show that varying
these parameters has a small impact on MRF electricity consumption (Pressley et al. 2015). The
sensitivity of GHG emissions to changes in the energy use of individual sorting equipment is
also very small compared to sensitivity to variations in the diesel consumption of trucks (Combs
2012). These findings demonstrate that, while it is important to include transportation and MRF
energy costs in a life cycle analysis, energy benefits from substitution of recycled material for
virgin material dominate the overall equation. In this analysis, therefore, we focus on material
recovery rates and material quality, and neglect the small variations in electricity consumption
associated with different sorting equipment.
An unanswered question in the literature is how variation in the sorting performance of
MRFs impacts downstream processes, and whether these variations change the results of waste
LCAs. Combs (2012) and Pressley (2015) use a linear model to quantify costs and emissions
associated with MRF operation for use in a solid waste management decision support tool,
similar to an LCA model. 1 While this work presents valuable information about MRF
operations, costs, and energy use, it includes only the mass of recovered material, and neglects
the fraction of incoming waste lost in the sorting process, and the quality of the material
recovered at an MRF. It also does not account for the impact that variations in input composition
and sorting efficiency have on the composition of MRF outputs. Furthermore, the separation
efficiencies assumed for MRF equipment are based on engineering intuition, rather than
experiment, and apply only to the target material for a given piece of sorting equipment.
Understanding how each piece of equipment sorts not only its target material, but also the other
components of the waste stream, is necessary to evaluate how variations in MRF parameters
impact the quantity and quality of recovered material. This work fills in some of the gaps in the
MRF literature by presenting experimentally-determined sorting efficiencies for MRF
equipment, including the sorting efficiency for non-target materials.

The decision support tool is the Solid Waste Optimization Life-Cycle Framework (SWOLF) developed by Levis
et al. (2013).

25
1.5 Outline for the thesis

The goal of this work is to apply model insights and case study data to examine the
recovery rate and quality of materials recovered by single stream MRFs in the United States.
We then compare results from our model with the assumptions used to evaluate the energy and
greenhouse gas impact of recycling in the Waste Reduction Model (WARM), the EPA's waste
management decisions support tool.
The remainder of this work is structured as follows: Chapter 2 details MRF operation
and the computational model, with a focus on the data obtained to support model development,
the calculation of experimental sorting efficiencies, and observed effects of input variation on
sorting performance. Experimental values are compared with those from the literature, and the
modeled results of several case studies presented. Chapter 3 explores the results from the
optimization function of the model to determine the best case MRF performance for single
stream recycling. Chapter 4 applies the MRF model results from Chapters 2 and 3 to waste
management LCA to assess the greenhouse gas impact of changing model assumptions to
reflect experimental data, and considers the policy implications of the model. Chapter 5 presents
the conclusions from the thesis.

26
Chapter 2
Modeling Material Recovery Facilities
A Material Recovery Facility (MRF) consists of a network of sorting equipment
connected by conveyor belts that transport material from one sorting step to another, and finally
to collection units. The function of an MRF is to separate an incoming mixture of materials into
its material fractions, which include recyclable materials that are sold to secondary processors
and recyclers, and residue that is sent to landfill. In addition to mechanized sorting steps,
personnel may be employed to hand sort materials that are difficult to target with a machine.
For example, a person may pick large cooking pots off of the incoming waste conveyor, as
large, heavy items can damage the sorting equipment. People also carry out negative sorting or
quality control by picking contaminants out of a sorted stream before it is collected. The
performance of an MRF is determined by factors including the size and type of equipment used,
the number of people employed, the arrangement of people and machines within the plant, and
the composition of the incoming material.
Although single stream recycling, which relies on MRFs to sort materials, is gaining
popularity in the United States, there is a lack of detailed information in the literature about the
sorting efficiency and output quality from MRFs (Cimpan, Rothmann, et al. 2015). As we
learned through collaboration with the operator of an MRF near Madrid, Spain, plant operators
also lack detailed knowledge of their facilities, and plants are designed and operated based on
rules of thumb, rather than site-specific data. The product of our work at the Spanish MRF was
a software model that can be customized to represent any sorting facility. The model allows the
user to evaluate the effects of changing plant parameters and input composition scenarios.
The objective of this chapter is to describe the operation of MRFs and the parameters
that determine the quantity and quality of recovered material. These parameters include the type
of materials that enter the plant, and the efficiency with which each sorting unit separates one
or more target materials from the mixed stream. Application of the MRF model, including
evaluation of single-stream recycling scenarios and possible improvements with optimized
MRF design, follow in Chapter 3.

27
2.1 Metrics of MRF Performance

The ability of an MRF to sort material is quantified using the metrics of recovery and
grade. Recovery is the fraction of one or many incoming desirable materials that are correctly
sorted, and grade is the purity of an output, or the fraction of the output that is composed of the
desired material. The materials that a sorting unit is designed or programmed to separate from
the mixed input are the target materials for that unit. Non-target materials are those that are not
targeted by the unit. For example, the target materials for a magnet sorting unit are ferrous
metals, and any other material that enters the magnet is a non-target material. Wolf (2011)
defines recovery and grade for systems of binary sorting steps, or those with two output streams.
In binary cases, each output can be identified as target, collecting the desirable material for the
sorting step, or secondary. Some sorting equipment have three or more output streams,
however, so we adopt a notation that can accommodate sorting units with any number of output
streams.

Output Unit 1

fil

f12

f31

/0
p2 Sorting Unit

f2i

f22
Input Stream
f2
"**' 0
Output Unit 2

Figure 2-1: Sorting step with two output units

28
Figure 2-1 illustrates a generic sorting unit. The input stream contains three materials
(M = 3), and waste is sorted into two output streams. If material m is the target material for
output i, then the recovery of material m into output i is:

M
2-1) Rm (
lI-m

where fib) is the mass of material m correctly sorted into output i, and Mm is the mass of
material m in the input to the sorting unit. The grade of the output is the mass of the target
material m in output i over the total mass of output i:

Gm = M (2-2)

The recovery and grade of materials in a system of sorting units is defined similarly. As
an example, a simple MRF is shown in Figure 2-2. The mixture of materials entering the plant
is the input stream. The waste is carried on conveyor belts, represented by arrows, through a
series of mechanical sorting steps, represented by rectangles. The arrow connecting step 2 back
to step 1 is called a recirculation loop. The output units, represented by ovals, collect the sorted
materials. In any realistic system, there will be some non-recyclable material in the input, so
one of the output units will be landfill. The QC step is a quality control station where one or
more personnel hand-pick contaminants out of the material stream before it is collected as an
output.

29
(:Input

Stng Stp

Sorting Step
4/
Sorting Step QC Output

te

Output Landfill
1

Figure 2-2: A simple MRF with three binary sorting units

The recovery of a material in this system is the same as for a unit process, except that

Pm is the mass of material m entering the system. For materials that are collected in multiple
output streams, the recovery is calculated by dividing the sum of correctly sorted material by
the input mass. For example, if material m is the target material for both outputs 1 and 2, then
the recovery of m in this system is:

Rm M f (2-3)
PM

where fl' is the mass of material m collected in output 1, and f2mis the mass of material m
collected in output 2.

30
2.2 Sorting Equipment in an MRF

This section describes the sorting equipment modeled in this thesis. A list of sorting
equipment and terminology is defined first, followed by a more detailed description of the
equipment studied in in depth for this work.

2.2.1 MRF Terminology and Sorting Equipment

Ballistic Separator: Several parallel inclined paddles move with an elliptical motion relative
to each other, causing material to bounce. Lightweight, flat materials such as paper and
cardboard are bounced towards the top end, while heavier, rolling objects such as metal cans
and plastic bottles roll downwards along the bouncing conveyors. Small particles fall through
holes in the paddles.
Disc Screen: Sort for paper and cardboard. Disc screens consist of an inclined plane with
parallel rods perpendicular to the direction of material flow. Discs are attached to the rods at
regular intervals. The spinning of the discs advances large, flat material forward up the incline,
and smaller items fall through the gaps between the rods. The size of the gaps between the rods,
and the angle of incline, determines the type of material captured by the disc screen. Screens
used to sort old corrugated cardboard (OCC), for example, have larger gaps, and those sorting
for paper have smaller gaps and a steeper incline.
Eddy Current: Eddy current separators target conductive nonferrous metals such as aluminum
and copper. A magnetic rotor with alternating polarity induces a current and magnetic field in
metals. The induced magnetic field opposes the inducing field, causing the metals to be repelled
over a splitter and into a collection unit.
Glass Breaker Screen: This is a screen positioned so that material falls from the preceding
conveyor onto the screen, causing glass containers to break. The glass pieces fall through the
screen, while other materials remain on top.
Over-belt Magnet: A magnet positioned above the conveyor and surrounded by a second
conveyor that moves attached ferrous materials into a region away from the magnet where they
fall into a collection unit.

31
Optical Sorters: General class of sensors that identify materials using optical technology
including cameras, lasers, and near-infrared (NIR) spectroscopy. Identified materials are
removed from the mixed waste stream by an air jet.
Quality Control: Personnel performing manual negative sorting of an output stream to remove
contamination.
Scalping Screens: Similar to a disc screen, but with smaller gaps between the rods, and
operated flat, rather than on an incline. Used to sort out small particles of debris (fines), usually
placed before the glass breaker screen to remove unwanted fines that would otherwise enter the
glass output.
Vacuum: A vacuum is used to remove plastic bags and other plastic films from the input
stream. It is suspended over the input conveyor, and guided over films by a worker. The
removed film may be sold, or diverted to landfill.

2.2.2 Principles of Separation for Select Equipment

Magnet

Ferrous metals are separated from other waste based on their magnetic properties. An
over-belt magnet is positioned above the mixed waste conveyor, as shown in Figure 2-3. The
magnetic force attracts ferrous metals, causing them to accelerate off of the lower conveyor and
to stick to the upper conveyor covering the magnet. Non-ferrous metals and other waste are
unaffected and remain on the lower conveyor. As the upper conveyor rotates, it carries the
ferrous particles away from the magnet so they can fall onto a third collection conveyor. Once
an over-belt magnet is installed, the operator can adjust its height above the conveyor, and the
speed of the upper conveyor belt.

32
am
*... Cie

-
Figure 2-3: Over belt magnetic separator (image source: BLS Magnet 2017)

Eddy Current Separator

An eddy current separator uses a magnetic rotor with magnets of alternating polarity to
induce an electric current in nonferrous conductive metals. The magnetic field in the ferrous
particles resulting from the eddy currents opposes the magnetic field of the rotor, and causes
the ferrous material to be repelled over a splitter as shown in Figure 2-4. Non-magnetic
materials, represented by squares in Figure 2-4, are unaffected by the rotating magnet, and
simply fall off the end of the conveyor.
The deflection force, Fd, acting on the particle is approximated as:

Fd = KBefm p (2-4)
P

where Be is the effective magnetic induction,fis the oscillation frequency of the magnetic field,
m, a and p are the mass, conductivity, and density of the ferrous particle, respectively, and K
and p are coefficients (Koermer 2015). To adjust the deflection force, operators can increase or
decrease the speed of the rotor to increase the magnetic field oscillation frequency.

Figure 2-4: Eddy Current Separator (image source: Vazzoler 2017)

33
Near Infrared Detector (NIR)

NIR detectors are a type of optical sorting technology. They are a more flexible sorting
technology than the magnet or eddy current, and allow the user control of settings through a
computer interface. Materials on a conveyor passing under the detector are identified by an NIR
camera, identified as the spectrometer scanner in Figure 2-5, which detects the near infrared
spectrum of the objects on the conveyor, and matches them against a library of known spectra
to identify the materials. If a material is one of those programmed for selection, then an air jet
is activated to blow the targeted material over a splitter and into the collection bin. Non-targeted
materials fall directly off the end of the conveyor into a collection unit.
0

o Feeding of unsorted material


Spectrometer scanner 0
Separation chamber

Figure 2-5: Near Infrared (NIR) Detector (image source: Tomra 2017)

NIR units may be programmed to select a wide range of polymers, and paper. The user
can adjust the selectivity and suppress selectivity settings, which control recovery and grade.
Selectivity adjusts the confidence with which the computer must identify an object before
activating the air jet. A higher selectivity requires a higher confidence of correct material
identification. This improves the grade of the output, since fewer non-target materials are
selected, but decreases recovery of the target material. Suppress selectivity allows the user to
avoid contamination with specified materials by suppressing the air jet if a specified non-target
material is identified next to a target material. For example, if an NIR is being used to recover
PET bottles, contamination of the output with other types of plastic, in particular PVC, is
problematic. If PVC is specified in the suppress selectivity setting, then if the computer detects

34
PVC near a PET bottle, it will suppress the air jet that would otherwise hit the PET bottle to
avoid accidentally blowing the PVC into the target output as well.

Ballistic Separator

A ballistic separator operates based on physical properties of size, shape, and density.
The separator uses parallel inclined paddles that move with an elliptical motion relative to
each other. This motion causes material on the inclined plane to bounce; lightweight, flat
materials such as paper and cardboard are bounced towards the top end, while heavier, rolling
objects such as metal cans and plastic bottles roll downwards (see
Figure 2-6). Small particles fall through the gaps between the paddles. Ballistic separators are
similar to disc screens, but have the advantage of separating materials into three outputs,
rather than two.

Rat &Rexae Items


Paper, OCC
& Aim

Rol"n w S~gdf
Rd IteoS Glass aed Trash
centahmes

Figure 2-6: Ballistic separator (image source: Tinsley Company 2017)

2.3 MRF Model and Unit Process Sorting Efficiency

The probability with which a sorting unit diverts a material into an output stream is the
sorting efficiency for that material and output. Using our definition for recovery, the sorting
efficiency of a target material into its target output is equal to that material's recovery. Sorting
efficiency is used to quantify the fraction of any material sorted into any output, regardless of
whether it is a target material or not. The sorting efficiency r for material m in output i is:

35
i=
fm (2-5)
Am

where fjis the mass of material m collected in unit I.


There is one sorting efficiency for each material in each output of a sorting unit. For
example, for the sorting unit in Figure 2-1, there are six sorting efficiencies, one for each of the
three materials in each of the two outputs. We assume that sorting units have no storage
capacity, so mass conservation requires that the sorting efficiencies for one material across all
output units sum to one, i.e., V rm = 1. For a binary sorting process, therefore, the sorting
efficiency of a material into one stream may be found by subtracting the sorting efficiency of
that same material into the other stream from 1. For example, if there is a 90% probability that
the magnet will correctly sort ferrous metals into the ferrous output, which we will identify as
output 1, then rjerrous = 0.9, and r2errous = 0.1.

Modeling an MRF requires knowledge of the sorting efficiency for each material at each
sorting step. The overall recovery for a facility could be determined by simply measuring the
input and output to the plant, but this approach would not provide the level of detail necessary
to model material flows within the plant, or to predict how changes to individual sorting steps
impact plant performance, primary goals of the modeling effort. There are two main challenges
to finding the sorting efficiency of equipment in an MRF. First, obtaining samples from within
an operating MRF requires interrupting normal operations. This is disruptive and costly, so
requires considerable cooperation and incentive on the part of the MRF operators. Second, to

calculate sorting efficiencies, the composition and mass of the input and outputs to each sorting
unit must be measured. Waste composition is determined by collecting samples and sorting
them by hand into their different material components. This is a very labor-intensive process,
and the representativeness of the samples is limited by the mass of waste that people can pick
through in a reasonable amount of time.
The challenges associated with calculating the sorting efficiency of equipment in an
MRF is reflected by a lack of experimental sorting efficiencies in the literature. Previous MRF
models rely on engineering intuition for sorting efficiency (Combs 2012; Pressley et al. 2015).
The estimated throughput and sorting efficiencies form one study of single stream MRFs in the
United States are shown in Table 2-1.

36
-1

Table 2-1: Estimated target material sorting efficiencies for single-stream MRF (Coombs, 2012)

Throughput Target Material Target Material


Sorting Equipment
(T/hr) Sorting Efficiency

Vacuum 10 Plastic film 0.98


Disc Screen (cardboard) 45 Corrugated cardboard 0.75
Scalping Screen 7 Fines 0.85
Glass Breaker Screen 9 Glass 0.97
Optical Glass 5 Colors of glass 0.98
Disc Screen (paper) 21 Paper 0.94
Optical PET 5 PET 0.98
Optical HDPE 5 HDPE 0.98
Magnet 2.25 Ferrous metals 0.98
Eddy Current 10 Aluminum 0.97
Manual Sorting 2 Various 0.98

These sorting efficiencies are a useful reference, as they are based on the practical
experience of MRF operators, however they only quantify the sorting efficiency for the target
material of each sorting step, and do not provide any information about the fate of non-target
materials. Target material recovery can be found for the MRF using these values, however the
following three important metrics cannot be calculated without also knowing the non-target
sorting efficiencies:
Grade
The grade of an output depends on the mass of the target material, and the total mass of
material collected in that output. For example, the target material for the vacuum in Table 2-1
is plastic film (including plastic bags), which are sorted with an efficiency of 0.98. The recovery
of plastic film is therefore 98%. The grade of the plastic film output, however, depends on how
many other materials are mistakenly vacuumed up along with the plastic. If the non-target
sorting rate is zero, then the only material in the plastic film output is plastic film itself, and the
grade is 100%. Given the mechanism of plastic film recovery, however, it is likely that some
fraction of other materials would be vacuumed up as well. The grade of an output material is
important for determining both its market value, and how it is used as a recycled material.

37
Contamination
In addition to allowing calculation of the total mass of the target output, and therefore
the grade, non-target sorting efficiencies determine which non-target materials are incorrectly
sorted into a target output. This is important because some materials are more problematic as
contaminants than others. Glass, for example, can wear down processing equipment, so
knowing whether the non-target materials in the plastic film output (i.e., the contaminants) are
glass fragments or, for example, paper, could determine the downstream use of the sorted plastic
film or how it is treated in a subsequent processing step.
Target material loss rates
The non-target materials in one sorting step may be target materials for a downstream
sorting unit. How well non-target materials are sorted determines the fraction of incoming
material to the system that reaches other target sorting units. For example, the vacuum is used
to remove plastic bags and other plastic films from the input waste stream. If the vacuum is the
first unit in a sequence of sorting steps and incorrectly sorts 5% of incoming paper into the
plastic film output, then only 95% of incoming paper reaches the disc screen. Neglecting any
intermediate sorting steps, the maximum recovery of paper by the system shown in Figure 2-7
is the product of the sorting efficiency of paper at each step:

Rpaper = (rTaper)(raper) = (0.95)(0.94) = 0.88

Output I Output 3
(Target: Plastic Film) (Target: Paper)

pap er 5 ap er

Mpaper Output 2 Disc Output 4

p2 screen 4
F ap er oaper

Figure 2-7: Sorting efficiency of paper in a two-step sorting process

38
2.3.1 Experimental Determination of Sorting Efficiencies for MRF Model

We developed a network flow model of an MRF to enable the evaluation a system's


sorting performance under different operating conditions, including different unit sorting
efficiencies, and to consider optimization of MRF design. The project was conducted in
collaboration with Spanish waste management company using a mixed waste MRF (Spanish
MRF) as a case study for data collection and model verification. The objective of the project
was to provide insight into MRF operating parameters and their relationships to MRF
performance to help operators improve plant operations and profit. The development of the
model and its computational structure is described by Testa (2015). Ip (2016) also examines the
uncertainty in input composition and conducts a sensitivity analysis of model parameters.
Determining the sorting efficiencies of the sorting units in the Spanish MRF was a key
component of developing the model. Thanks to the enthusiastic cooperation of MRF operators
and management, we had the opportunity to collect data from the MRF at a level of detail
unavailable in the literature. Many of the equipment used in the Spanish MRF are common to
sorting operations in the United States and elsewhere, so we can apply the sorting efficiencies
calculated based on our experimental data to model other MRFs. There is variation in sorting
efficiency between equipment types, operational settings, and input stream composition,
discussed in greater detail in Section 2.5, so application of our case study results to other MRFs
involves some uncertainty. The experimental sorting efficiencies however, are an improvement
over assumed efficiencies, and provide information about the flow of all materials, target and
non-target, in the MRF.
The Spanish MRF is a mixed waste sorting facility, so the input contains a wide variety
of materials, and some of the sorting equipment is designed to remove organic matter and other
materials that are not commonly found in single stream recycling. A schematic of the entire
mixed waste MRF is included in the Appendix. One section of the Spanish MRF, however, is
focused on collecting packaging waste, and is therefore similar in design to a single stream
MRF. This section, known as the Light Packaging Recovery Section (LPRS), is shown in Figure
2-8, and was the focus of our work to measure the sorting efficiencies of equipment.
The input stream to the LPRS is the output from other parts of the plant. Upstream
sorting processes remove organic matter, bulky waste, glass, paper and cardboard, so the input
stream to the LPRS is similar in composition to single stream recycling, but without paper or

39
glass. The paper and cardboard are sorted elsewhere in the plant, so we were able to obtain
experiential sorting efficiencies for those materials as well. We note that Tetra Pak2 is the brand
name of aseptic beverage cartons popular in Spain and other European countries.

from
Plant
CInput

S MagnetFeru

NIR 1

PET
NIR2 NR3
Output

NIR4 NIR 5 Patc


Output

Recirculation: Plastics, Metals, TetraPak


NIR 6Eddy Current

Landfill

Figure 2-8: Light packaging recovery section (LPRS) of Spanish MRF

Sorting efficiencies for each piece of equipment in the LPRS were determined based on
the composition of the input, output, and recirculation streams. Table 2-2 shows five different
measurements of the composition of the input to the magnet, the first sorting step of the LPRS.
The composition for data set 1 were determined by directly sampling the input to the magnet;
the other data sets are based on combinations of measurements of the outputs of upstream
sorting equipment that precede the magnet in the plant.

2 See http://www.tetrapak.com/us for more information about the product.

40
-J

Table 2-2: Composition of the input to the Light Packaging Recovery Section magnet for five different
samples

Materis Composition by Mass


Data Set 1 2 3 4 5 Average
Aluminum 0.01 0.01 0.01 0.01 0.01 0.01
Ferrous 0.13 0.12 0.09 0.12 0.11 0.11
Glass 0.00 0.00 0.01 0.00 0.00 0.00
HDPE (High Density Polyethylene) 0.03 0.04 0.03 0.04 0.03 0.03
Organic 0.22 0.07 0.06 0.04 0.04 0.09
Other 0.28 0.31 0.39 0.29 0.44 0.35
Other Plastics 0.00 0.01 0.01 0.01 0.00 0.01
Paper & Cardboard 0.06 0.09 0.07 0.27 0.14 0.14
PET (Polyethylene terephthalate) 0.09 0.09 0.11 0.11 0.11 0.11
Plastic Film 0.09 0.16 0.11 0.06 0.05 0.08
Polypropylene 0.04 0.05 0.07 0.03 0.03 0.04
TetraPak 0.04 0.05 0.04 0.04 0.03 0.04

The output streams in Figure 2-8 were sampled and characterized by material type on
two different dates. The recirculation stream was also sampled, and a mass balance procedure
performed to determine the throughput in each piece of equipment. With a complete set of input
and output data, we back-calculated the input and output composition for each sorting step, and
calculated the sorting efficiency for each step. An improved estimation of the sorting
efficiencies for the LPRS was found by computing the sorting efficiencies that minimized the
weighted sum of the absolute errors between two output stream measurements and the modeled
output stream compositions (Testa 2015).
The resulting sorting efficiencies used in the model and as experimental values in this
thesis are shown in Table 2-3. For simplicity, only the sorting efficiency for the target output
stream is listed for each sorting unit. Since all of the sorting equipment in the LRPS may be
modeled as binary units, the sorting efficiency for each material into the non-target output may
be calculated by subtracting the given value from one. Target materials sorting efficiencies are
indicated in bold. For example, the target material for the magnet, ferrous metal, is sorted into
the ferrous output with efficiency 0.92. Non-target materials that were found to have a high rate
of incorrect sorting are highlighted by a grey background, for example 22% of glass is
incorrectly sorted into the PET and Tetra Pak output stream by NIR 1.

41
Table 2-3: Experimental separation efficiencies for LPRS sorting equipment

1 NIR 2 NIR 3 NIR 4 NIRS Current NIR 6


Equipment Magnet NIR
tP E E
uT, p e M
i xedCr et Pl ast i cs

,
Target Materials Ferrous TetraPak PET HDPE TetraPak Aluminum

Aluminum 0.00 0.05 0.06 0.00 0.03 0.02 0.80 0.25


Ferrous 0.92 0.00 0.00 0.00 0.00 0.01 0.00 0.17
Glass 0.00 0.22 0.00 0.00 0.00 0.18 0.76 1.00
HDPE 0.00 0.00 0.29 0.71 0.08 0.01 0.00 0.70
Organic 0.00 0.00 0.00 0.00 0.17 0.00 0.00 0.00
Other 0.02 0.01 0.20 0.00 0.03 0.05 0.00 0.01
Other Plastics 0.00 0.07 0.11 0.01 0.09 0.74 0.00 0.93
Paper & Cardboard 0.00 0.04 0.06 0.00 0.15 0.00 0.00 0.03
PET 0.00 0.86 0.93 0.00 0.01 0.01 0.00 0.93
Plastic Film 0.01 0.04 0.35 0.07 0.21 0.22 0.00 0.28
Polypropylene 0.00 0.06 0.11 0.00 0.05 0.73 0.01 0.86
TetraPak 0.00 0.71 0.04 0.00 0.96 0.00 0.04 0.92

The experimental sorting efficiencies above present a more complex picture of material
sorting than the estimated efficiencies in Table 2-1. First, the sorting efficiencies for target
materials at the Spanish MRF are relatively low, but this reflects the role of people in quality
control. The estimated efficiencies in Table 2-1 assume that personnel are used as quality
control after the mechanical units to achieve the given sorting efficiencies. The LPRS data, in
contrast, provides insight into how the equipment alone performs, which is more useful when
considering possible configurations of sorting equipment. Personnel are used for quality control
at the Spanish MRF when required to improve the grade of an output, but they were not active
during the tests to determine sorting efficiencies.
Second, the data shows that there are some materials that have a higher rate of incorrect
sorting, and are more likely to become contaminants. The impact of incorrect sorting on output
composition and grade depends strongly on the composition of the input. For example, if the
amount of glass entering the LPRS is very small relative to the other materials, in particular
PET and Tetra Pak, then mass of the glass that contaminates the PET and Tetra Pak output will
also be very small, and the grade of the output high. If, however, the amount of glass in the
input is comparable to PET and Tetra Pak, then the grade of the output from NIRl will be low.
Knowing the input composition is therefore the next step in modeling an MRF.

42
2.4 MRF Input Composition

The composition of material entering the sorting facility is a user inputs to the model.
The breakdown of materials within the input stream, and the rate at which it enters the MRF,
determines the total recoverable material. It also determines the amount and nature of non-
recyclables that can hinder the sorting process and become contaminants in the output streams.
Unfortunately, the composition of the material that enters MRFs is not well defined. This
section describes some of the challenges associated with obtaining waste composition data, and
presents literature values to compare with the experimental data obtained from the Spanish
MRF.
In the United States, the EPA reports the materials generated in MSW every year. Table
2-4 summarizes the total material generated as MSW in the United States in 2005, 2010, and
2014, as well as the fraction of each material category that was recycled each year.

Table 2-4: Materials in generated and recycled in the United States in 2005, 2010, and 2014 (U.S. EPA
2016b)

Materials Generated in MSW Materials Recycled


(Thousands of metric tons) (Percent of generation)

Materials 2005 2010 2014 2005 ] 2010 2014


Paper and Paperboard 76,987 64,710 62,260 49.5% 62.5% 64.7%
Glass 11,379 10,454 10,417 20.7% 27.2% 26.0%
Ferrous 13,802 15,354 16,053 33.0% 34.3% 33.0%
Aluminum 3,022 3,185 3,203 20.7% 19.4% 19.8%
1 66.7%
IOther Nonferrous 1,688 1,833 1,85 68.8% 71.3%
Plastics 26,661 28,494 30,172 6.1% 8.0% 9.5%
Rubber and Leather 6,615 7,033 7,450 14.4% 18.6% 17.5%
Textiles 10,445 11,996 14,719 15.9% 15.5% 16.2%
Wood 13,421 14,256 14,628 12.4% 14.5% 15.9%
Other 3,893 4,274 4,029 28.2% 29.1% 29.1%
Food 29,882 32,432 34,846 2.1% 2.7% 5.1%
Yard Trimmings 29,102 30,309 31,307 61.9% 57.5% 61.1%
Misc. Inorganic Waste 3,348 3,485 3,603
T4otal 230,245 227,813 234,537[ 7 31.4% 34.0% 34.6%

43
The data for generated materials includes all nonhazardous materials that enter the waste
management stream excluding pre-consumer industrial scrap. The EPA data for recycled
materials is the sum of reported purchases of post-consumer collected materials and net exports
of the material. Thus, the amount of recycled material listed accounts for the outputs from
sorting operations, but does not include MSW that is collected for recycling but lost during the
sorting process. For example, we measured the recovery of ferrous material by the magnet in
the Spanish MRF as 92% (see Table 2-3). If we neglect the recirculation loop, then the
maximum fraction of ferrous metals recovered by the LPRS is 92%, and the remaining 8% of
incoming ferrous metals are either sent to landfill, or become contaminants in other output
streams. If we assume that this sorting efficiency is representative of magnets in general, then
the total ferrous material reported as recycled by the EPA is only 92% of the ferrous material
collected for recycling. Using data from 2014, and assuming a 92% recovery rate of ferrous
metals in sorting processes, 5297.5 thousand metric tons of ferrous material were sold, but 5758
thousand metric tons were collected for recycling and would have factored into the composition
of waste entering an MRF.

2.4.1 Characterization of Input Material

One of the reasons that accurate waste composition data of direct inputs to MRFs is
scarce is that obtaining detailed information about the composition of inputs and outputs to
MRFs is costly and time-consuming. The author was able to observe (but fortunately not to
conduct herself) the process of obtaining waste characterization measurements at the Spanish
MRF as part of the model development process. The process highlights the challenges of
obtaining reliable, representative waste composition data, and introduces sources of uncertainty
in that data.
The best way to characterize the composition of inputs and outputs to an MRF is to sort
samples by hand. Mechanized sorting cannot be used because equipment within an MRF does
not sort perfectly, and one of the objectives of characterizing the input waste is to calculate the
sorting efficiencies of equipment. Samples of waste are spread out on a table and sorted into
bins by material. Once completely sorted, each bin is weighed, and the final composition
determined. The process is unpleasant, as recyclables often contain some food waste and other
organic matter that decomposes. As the Spanish MRF processes mixed waste, the inputs are

44
very dirty, but there is some level of contamination in all recycling streams, as many of the
containers collected for recycling were originally used to contain food. Hand sorting also
requires considerable expertise and judgement. For example, while intact plastic bottles are
easy to identify by their labeling, fragments of plastic material such as portions of bottles or
other packaging require knowledge of the properties and common uses of different types of
polymers.
Another challenge to sampling waste, other than the time it takes, is obtaining a
representative sample. MSW is processed at throughput rates of around 30 T/hr (Combs 2012),
but it takes several hours for people to sort through a 200kg sample. Furthermore, the waste is
heterogeneous, and may experience sifting or stratification during transport so that smaller
particles like glass fragments and organic matter fall to the bottom of a large container, while
larger pieces float on top. The Spanish MRF used the "quarters" procedure, shown in Figure
2-9, to obtain as representative a sample as possible. A large mass of waste is assembled from
different parts of a waste bunker (e.g., some material from the top, middle, and bottom of the
container) and, using a front-end loader, mixed and then divided into four sections. Two of
these sections are discarded, and the other two are combined to form a new pile, which is again
mixed and divided into quarters. This process is repeated until the remaining waste is 150

-
250kg, which is then carefully sorted by hand.

combine

discard

Figure 2-9: Quarters procedure for sampling heterogeneous waste

The MRF in Spain processes about 15,000 tons of incoming waste each month, and
employs an outside company to carry out characterizations of the plant input every three
months. The characterization often involves three samples. If each sample is about 200kg, the

45
total mass of input waste characterized over a three month period is about 0.001% of the total
mass of material processed by the MRF during that time. The composition of incoming material
is therefore a source of uncertainty in the model, and improvements in MRF modeling depend
in part on better means for quantifying the composition of waste.

2.4.2 Composition of Single Stream Recycling in the United States

Unlike the Spanish MRF, which processes mix waste and therefore has a high level of
organic matter and non-recyclable materials in the input, most MRFs in the United States
process single stream recycling. Two reports provide single-stream input composition data
obtained through hand-sorting in two different states (Cimpan, Maul, et al. 2015). The
residential single-stream data from each state is summarized in Table 2-5, with EPA data for
the United States in 2010 for comparison. The Pennsylvania data is the average of 45 samples,
each 113-136 kg, collected from the input to a single stream MRF in York, PA (R.W. Beck Inc.
2005). The Oregon data is the average of 47 samples of about the same size (136kg) from the
single stream input to 5 different MRF facilities in the Portland area (Oregon Department of
Environmental Quality 2011). Note that the material categories used by each report are slightly
different, as are the materials collected for recycling in the two states. The greatest difference
reflected by the data is that glass is collected in Pennsylvania's single stream curbside recycling,
while in Oregon it is collected separately, and therefore not included in this data. The EPA data
is for the entire United States, and includes commercial as well as residential sources, which
contributes to the elevated amount of corrugated cardboard.

46
Table 2-5: Composition of residential single stream recycling (Combs 2012; R.W. Beck Inc. 2005;
Oregon Department of Environmental Quality 2011)

Average Composition by Mass

Materials Pennsylvania Oregon United States

Newspapers 44.1% 53.1% 13.5%


Other Paper 11.5% 13.4% 15.3%
Corrugated Cardboard 1.3% 18.3% 39.8%
PET Bottles 4.8% - 3.7%

HDPE Bottles 6.7% - 3.1%

Mixed Plastic 6.7% 5.4%


Glass 20.6% 1.4% 12.8%
Steel Cans 5.4% 2.1% 3.8%
Aluminum Cans 1.9% 0.3% 2.0%

Aluminum Other 0.2% 2.1% 0.6%

Non-recyclable Materials 3.6% 2.6%

2.4.3 Variation in Waste Composition

Waste stream composition is variable, so data from one community may not be
generalizable to others. Variations are significant not only between states, but also between
communities in the same states, and over time in the same location. Figure 2-10 shows variation
in the amount of three categories of materials in the samples taken at the Pennsylvania MRF.
The waste samples originated in 8 different communities or transfer stations, which are
identified on the chart as sources A-H. The material categories are paper, containers (plastic
bottles and aluminum cans), and rejects. Rejects or residue are the components of waste that
enter an MRF but cannot be recycled. They could include almost any material category,
including, for example, food waste, Styrofoam, and recyclable materials that are dirty or wet.
. Variation was observed in the composition of collected material even between
communities that use the same collection system guidelines for what residents can and cannot
recycle. Possible reasons for these variations include local education efforts about recycling
and what materials are allowed. Conversation with the operators of an MRF in Massachusetts

47
support this, as they cite consumer education as an important part of obtaining good-quality
MRF inputs. Other reasons for the variations in the PA data include differences in newspaper
subscription rates in urban and rural areas, regional differences in available products (e.g., the
brand of milk available in the local store could impact the types of plastics that people put into
recycling), and relative levels of affluence, which help determine consumer choices for products
(R.W. Beck Inc. 2005).

80%

70%

60%

50%

40%

30%

20%

10%

U'%o
---- --- ---- Eu.
-

Paper Containers Rejects

*SourceA ESourceB ESourceC ESourceD SourceE ESourceF ESourceG *SourceH

Figure 2-10: Variations in composition of input waste at a single stream MRF in York, PA (R.W. Beck
Inc. 2005)

Data collected at the Spanish MRF demonstrates variation in input composition over
time. Figure 2-11 shows the fraction of eight different valuable materials in entering the LPRS
of the Spanish MRF for five different sample dates (data sets 1-5 in Table 2-2). Note that as
this is a mixed-waste MRF, there were also a large number of contaminants in the input.
Materials including organic matter and other non-recyclable waste made up between 30% and
50% of the total input to the LPRS.

48
30%

25%

20%

15%

10%

5%

0%
-

Paper& Cardboard PET HDPE Film Polyproplyene Tetrapak Ferrous Aluminum

Material Type

Figure 2-11: Variations in the recyclable component of mixed waste input to a section of the Spanish
MRF over five sets of sample data. The first data set is the average of 5 samples taken on different days
over the course of a year. The others represent one sample each.

The data in Figure 2-10 and Figure 2-11 highlight the variability of input composition
to MRFs both as a function of source location, and over time. This variability limits the
application of national or regional average waste composition to local waste management
modeling, and argues for the need for better waste composition monitoring. Given the expense
of characterizing waste by hand, future improvements in MRF models as decision tools for
MRF operators, or local waste managers, will depend on innovations in automated waste
characterization technology.

2.5 Variation in Sorting Efficiency

The sorting efficiencies used in the model are assumed to be independent of input
composition and throughput (Testa 2015). Our experimental sorting efficiencies, however,
reflect the performance of the system under one set of conditions, including the input

49
composition, throughput, and other operating parameters, and may not be the same under
different conditions (Wolf 2011). Intuitively, there should be some variation of sorting
efficiency with feed rate (throughput) and the mixture of materials entering a sorting unit, but
literature data on how variations in these parameters effect sorting efficiency under realistic
sorting conditions is unavailable.
Wolf (2011) investigates the effect of input material composition on the performance of
three types of separation equipment under experimental conditions. Two of these technologies,
the eddy current separator and a simulated over-belt magnet, are used in the Spanish MRF and
common to single-stream recycling. Wolf finds that for eddy current sorting of a mixture of 2-
inch squares of low density polyethylene (LDPE) and aluminum, increasing the aluminum
concentration leads to a higher rate of LDPE incorrectly sorted into the aluminum output
stream, while the sorting efficiency for aluminum into its target output remains unchanged. In
other words, increasing the aluminum concentration causes a decrease in LDPE sorting
efficiency. Wolf suggests this phenomenon is due to overlap of the aluminum and LDPE
squares so that the aluminum pieces carry the plastic squares over the divider into the aluminum
collection bin.
Using a simulated over belt magnet, Wolf finds that both target and non-target
separation efficiencies are dependent on concentration. The experiment was performed with a
mixture of chips of either iron or steel and aluminum, and showed that interference between the
two types of particles decreased the sorting efficiency for both. This could be caused, for
example, by a layer of aluminum chips weighing down a steel chip so it cannot attach to the
magnet, or, conversely, by steel chips carrying aluminum chips to the magnet.
While these experimental results are not immediately generalizable to real MRF
conditions, where material mixtures are complex and the shapes of objects varied, they do
demonstrate that changing the concentration and shape of materials in a mixture can effect
sorting efficiency. Wolf also shows mathematically how changing the splitter position, a
parameter of the equipment, alters the distribution of material collected in the target and non-
target outputs.

50
2.5.1 Design of Experiments to Test Sorting Efficiency Variation

We conducted experiments to quantify the magnitude and direction of the effect of


changing input composition and equipment parameters on sorting efficiency for three types of
sorting equipment using realistic input samples. The sorting technologies tested were the
magnet, eddy current, and a near-infrared (NIR) unit. These were selected because they are the
sorting technologies used in the light packaging section of the Spanish MRF, and are similarly
important in single-stream recycling as they capture ferrous metals, aluminum, plastics, and, in
some cases, paper, all valuable outputs from a sorting facility. Each of these three technologies
separates objects based on material properties, and each has one or more parameters that the
user can adjust. The magnet and eddy current were tested at the Spanish MRF by plant operators
and staff. To minimize disruptions to the plant, and to have more control over the testing
conditions, we ran our NIR experiments at a testing facility run by the manufacturer of the NIR
units used in the Spanish plant. We set the test NIR to target PET so it would model the
performance of NIR 2 in the LPRS of the Spanish MRF (see Figure 2-8).
Two tests were carried out on each of the three sorting units. For the first test, the
equipment parameters were held constant while the concentration of target material in the input
was varied. For the magnet and eddy current, we varied the concentration of ferrous and
aluminum, respectively, between low, medium, and high. The medium concentration was equal
to the concentration measured at the Spanish MRF. For the NIR, we again used one sample that
represented the normal input concentration, based on measurements at the Spanish facility, and
chose four other sample compositions to represent different scenarios. The second test for each
sorting unit involved holding the sample composition constant, and varying one equipment
parameter. The parameters were the height of the magnet, the rotation speed of the eddy current
rotor, and the selectivity settings of the NIR. The samples for the magnet and eddy current were
about 150kg each, while the samples for the NIR units were smaller, about 25kg each. We used
smaller samples for the NIR experiment because they were shipped from the Spanish MRF,
where they were prepared, to the testing facility near Frankfurt, Germany.
The test procedure was to run a sample through the equipment, and then characterize
the two output streams using hand-sorting. For the magnet and eddy-current, materials were
characterized simply as target or non-target. For the smaller samples used in the NIR tests, it
was feasible to sort the output streams into their individual material types.

51
An example of the data obtained from the magnet test is shown in Table 2-6, and the
resulting sorting efficiencies for the two material categories are in Table 2-7. This was the
baseline test for the magnet, with the normal concentration of ferrous material in the input
(about 15%) and the usual height of the magnet (350mm above the conveyor). The recovery
and grade for this test were calculated as follows:

Rferrous - 2 0.89
=2kg

2 0.2 kg
Gferrous = 20.kg= 0.98
s 20.6kg

Table 2-6: Magnet Test 1.1 (normal ferrous input concentration and magnet height)

Material Stream Mass Description


(kg)
Input 150.8 Total material processed by the magnet
Ferrous material in input 22.5 Ferrous material processed by the magnet
u .a . Non-ferrous material processed by the
Non-ferrous material i nput 128.3 magnet
Target Output 20.6
-

Ferrous material in target output 20.2 Correctly sorted ferrous


Non-ferrous material in target output 0.4 Incorrectly sorted non-ferrous
Non-Target Output 130.2
Ferrous material in non-target output 2.3 Incorrectly sorted ferrous
Non-ferrous material in non-target 127.9 Correctly sorted non-ferrous
output

Table 2-7: Sorting efficiencies for ferrous and non-ferrous materials in Magnet Test 1.1

Non-Target
Material Target Output Output

Ferrous 0.898 0.102


Non-ferrous 0.003 0.997

52
__________

2.5.2 Magnet Test Results

The recovery and grade of the ferrous output from the magnet tests are plotted in Figure
2-12. The data show an inverse relationship between ferrous recovery and input ferrous
concentration (indicated as [Fe]). The grade is less variable. This suggests that the amount of
incorrectly sorted non-ferrous is proportional to the ferrous sorting efficiency. As more ferrous
material is correctly sorted, more non-ferrous material is incorrectly sorted into the same output,
so that the ratio between the two, grade, remains nearly constant. The chart in Figure 2-12 also
shows an inverse relationship between magnet height and recovery, indicated by the outlined
shapes on the chart, which would suggest that lowering the magnet height would improve
recovery. The operators noted, however, that the lower magnet height used for this test is not
feasible in practice, as large objects would become caught in the space between the conveyor
and the magnet.

1.0 U 0 X
0
0.9 A

0.8

0.7 - 0 Normal [Fe]

o 0.6 - High [Fe] (pressed)

0.5 *High [Fe]


U-
A Low [Fe] (pressed)
0.4
XLow [Fe]
w 0.3 ONormal Height

0.2 OGreater Height

0.1 A Lower Height

0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Recovery of Ferrous Metals

Figure 2-12: Recovery and grade of ferrous metals under varied target material concentration and
magnet height (experimental data from Spanish MRF)

53
This experiment also revealed that object geometry has a greater effect on recovery than
either material concentration or magnet height. The steel cans added to the input sample to
adjust the ferrous concentration were initially obtained from the baled ferrous output at the
MRF. When material is baled, it is compressed, and then bound, so the metal cans used in these
samples were flattened, whereas most of the tin cans in the normal input to the magnet are in
their original, cylindrical shape (see Figure 2-13). The tests that used the flattened cans are
labeled High [Fe] (pressed) and Low [Fe] (pressed) in Figure 2-12. The MRF employees
conducting the tests were surprised by the sharply reduced recovery rates of these samples. The
recovery rate for the sample with a low ferrous concentration and pressed cans is particularly
low at only 17%.

Figure 2-13: Cylindrical and pressed soda cans

The tests were repeated with pressed metal cans replaced by intact cans from the input
waste stream. These second tests are indicated by High [Fe] and Low [Fe] on the chart, and
show a dramatic improvement in the recovery for the sample with low ferrous material
concentration. This result suggests that if ferrous material is covered by other, non-magnetic
material, a condition made more likely if the cans are flattened, then it is less able to be
recovered by the magnet. This finding agrees with that of Wolf (2011), whose experiments with
aluminum and steel chips showed that the sorting efficiency for the steel decreased with lower
steel concentration. Wolf attributed this effect to entrapment of the steel chips when the amount
of steel was low.
Unlike in Wolf, the sorting efficiency of the non-target material showed little variation
with ferrous material concentration, as shown in Figure 2-14, and, with the exception of one

54
test, was constant at close to 1.0. The non-ferrous material in this test was representative of
MSW, and contained a variety of materials of different shapes and sizes. There was, therefore,
no consistent interaction between the non-ferrous materials and the ferrous that would create a
trend. The upper limit of ferrous concentration in the input samples used for these tests was
lower than in Wolf.

1.0 U U
S M

0.8

0.6
0
CL

a) 0.4

0.2
0

0.0
-

0.00 0.05 0.10 0.15 0.20 0.25 0.30

Input Concentration of Ferrous

0 Ferrous U Non-Ferrous OFerrous (Pressed Fe) 0 Non-Ferrous (Pressed Fe)

Figure 2-14: Separation efficiency for ferrous and non-ferrous material versus concentration of ferrous
in sample (experimental data from Spanish MRF)

2.5.3 Eddy Current Test Results

The results for the eddy current tests, presented in Figure 2-15, show that, as in the case
of the magnet, decreasing the target material concentration improved the recovery. Unlike the
magnet, however, where the normal operating height was the best feasible equipment setting,
the eddy current tests suggest that the grade of the aluminum output could be improved by
increasing the rotor speed. This finding may be explained by an increase in the deflection force
acting on the aluminum, causing them to accelerate faster away from other materials. The low
recovery and grade of aluminum under normal operating conditions suggest that improvements

55
in this step might be beneficial for the plant. The Spanish MRF currently employs personnel to
work as quality control on the aluminum output line to remove contaminants.

1.0 - 0 0

0.9 -AA
*0 El X
0.8
-

0 Normal [All
0.7
-

* High [Al] 0
E 0.6
XLow [Al]
E 0.5
.2 3 Normal Speed
0.4
0 High Speed
0.3
A Low Speed
0.2

0.1 1

0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

Recovery of Aluminum

Figure 2-15: Recovery and grade of aluminum output under varied aluminum concentration and rotation
speed of the eddy current magnet (experimental data from Spanish MRF)

2.5.4 Near Infrared Detector Test Results

The NIR units installed at the Spanish MRF are Titech Autosort, manufactured by
Tomra. Tomra operates an equipment testing facility in Koblenz, Germany, where we carried
out our experiments. We configured the test NIR settings to match those of the NIR at the
Spanish MRF used to select PET. For the first test, we prepared five, 25kg waste samples with
different amounts of PET and other materials, ran each of the samples through the test NIR, and
separated the output into material fractions. For the second test, we ran one of the samples, with
input representative of the average at the Spanish MRF, through the test equipment, but varied
the selectivity settings to target recovery or grade.

56
Varied Input Composition

Unlike the magnet and eddy current, the NIR can be programmed to select, or suppress
selection, of a range of materials. It was therefore necessary to track not only the target material
(PET in this case), but the other material categories as well. The composition of the five input
samples is shown in Table 2-8. Sample IA represents the measured composition of input to the
NIR unit that sorts PET at the Spanish MRF. In the LPRS at the Spanish MRF, the PET selection
step follows the magnet, which removes ferrous metals, and an NIR unit that diverts both PET
and Tetra Pak into the PET-selecting NIR (see Figure 2-8). The IA input therefore consists
mainly of PET and Tetra Pak. lB is the average input composition to all of the NIR units in the
LPRS. Samples IC and ID contain a high fraction of the valuable materials recovered in the
LPRS, other than ferrous, and IE contains elevated amounts of paper, ferrous, and HDPE.

Table 2-8: Composition of five input samples for NIR Test 1 (25kg each)

Fraction of Materials in Sample


Sample ID A B C D E
Aluminum 2.1% 0.9% 16.2% 15.9% 7.9%
Ferrous 0.0% 0.0% 0.0% 0.0% 7.5%
Glass 0.0% 0.0% 0.0% 0.0% 0.0%
HDPE 0.0% 2.5% 0.0% 11.1% 8.7%
Organic 0.0% 0.1% 0.5% 0.3% 1.0%
Other 3.2% 43.2% 34.9% 10.3% 14.9%
Other Plastics 0.1% 0.4% 1.4% 0.3% 0.8%
Paper & Cardboard 3.6% 15.7% 7.3% 3.4% 8.6%
PET 62.1% 18.0% 8.4% 12.2% 7.9%
Plastic Film 3.8% 0.1% 13.0% 16.6% 20.4%
Polypropylene 2.0% 2.5% 10.7% 18.1% 14.7%
Tetra Pak 23.1% 16.5% 7.6% 11.7% 7.4%

The recovery and grade of PET for each sample are shown in Figure 2-16, with the input
fraction of PET on the secondary axis. The recovery of PET is relatively stable, from a
minimum of 92% for sample A, to a maximum of 98% for sample C. The data suggest that
there may be some inverse relationship between PET concentration and recovery, but it is not
clear. The differences in recovery may also be due to the presence or absence of different

57
materials that hinder the recovery of PET, or to differences in the size and shape of the PET
bottles in each sample. As these samples were constructed out of real MSW, they reflect the
heterogeneity of a mixed waste stream, and therefore realistic conditions, but the variety of
material sizes, shapes, and colors adds to the complexity of the results. A useful conclusion
from these data is that the sorting efficiency of the NIR for PET is relatively insensitive to input
composition, as none of the samples had a dramatic decrease in efficiency.

1.00 0.99 1.0


0.98 0.98
0.97
5--0.95 09

0.95 -0-- -- - 0.7


- 08

SZ
0..
- 0.4
0.8 - 006- .5
0 .85 - - 0 .5

0.4 .s
S0.80
0.3
C
0
-50.2

S 0.1

0.70 - 0.0
A B C D E
Sample ID
Recovery of PET Grade of PET Output -N-PET in Input Sample

Figure 2-16: Results of input sample variation NIR test: recovery of PET, grade of target output, and
fraction of PET in input sample (experimental data from Tomra testing facility)

The breakdown by material type of the contamination in each output (see Figure 2-17)
reveals that some materials are more likely to be incorrectly sorted by the NIR than others. A
higher amount of these materials in the input leads to a lower grade of the PET output. Film,
aluminum, ferrous metals, and paper and cardboard are the primary contaminants in the PET
output. Film (mostly plastic bags) and paper pose particular challenges for this type of sorting
technology, and their light weight and large surface area make them easily caught in the air jets,
causing them to float into the target output.

58
12%

10%
Aluminum

8%

0
6%
0

Ferrous
4%

Aluminum Aluminum

2% TetraPak
Paper & Cardboard Po ypropy en Polypropylene Paper & Cardboard
Fin -~p PolypropyleneY
Mixe d Plastics
-fsnes
0%
A B C D E
SAMPLE ID

Figure 2-17: Composition of contamination in PET outputs for five different input samples
(experimental data from Tomra testing facility)

Another way to view the results from the NIR tests is through the sorting efficiencies of
materials into the PET output stream. For PET, this is the same as the recovery shown in Figure
2-16. The sorting efficiencies help explain the data in Figure 2-17 when compared with the
input composition of each sample. For sample A, for example, the sorting efficiency for Tetra
Pak is small, but Tetra Pak is 20% of the input mass, so it appears as a contaminant. The reverse
is true for plastic film, which composes only 4% of the input but is sorted into the PET output
with 23% efficiency.

59
Table 2-9: Sorting efficiencies for materials in target output stream of NIR unit selecting PET
(experimental data)

Sorting Efficiency in Target Output


Sample ID A B C D E

Aluminum 0.03 0.00 0.00 0.03 0.00


Ferrous - - - - 0.01
Glass - - -

-
HDPE - 0.00 - 0.00 0.00
Organic - 0.00 0.00 0.00 0.00
Other 0.11 0.02 0.01 0.02 0.02
Other Plastics 0.00 0.00 0.00 0.08 0.00
Paper & Cardboard 0.00 0.00 0.00 0.00 0.01
PET 0.92 0.94 0.98 0.95 0.92
Plastic Film 0.23 0.00 0.01 0.05 0.01
Polypropylene 0.00 0.00 0.00 0.00 0.00
Tetra Pak 0.01 0.00 0.00 0.00 0.00

Varied Selectivity Settings

Changing the selectivity settings of the NIR demonstrated the degree of control the user
has with an NIR unit. As Figure 2-18 shows, a lower selectivity for PET led to a higher recovery,
but also a decrease in grade as more contaminants were incorrectly sorted into the target output.
Conversely, increasing the selectivity reduced the recovery, but yielded a slight improvement
in grade. The plant managers concluded from this data is that the settings in use at the Spanish
MRF are a good compromise between recovery and grade. If we assume that these results can
be extended to sorting of all materials by NIR technology, then the possible design options for
a series of NIR units is very large. Our MRF model uses a fixed value for the sorting efficiency
of each unit, but future improvements could include adding a variable for selectivity.

60
11bo A

1.00

0.95

0.90

0.85

0.80

0.75

0.70
Normal Settings Lower Selectivity Higher Selectivity

E Recovery PET E Grade PET

Figure 2-18: Recovery and grade of PET output from test NIR using sample A (normal input
composition) and varied selectivity settings

The data presented in this section demonstrate that sorting efficiencies are not constant,
but rather vary with both input composition and equipment parameters. We did not measure the
impact of throughput on sorting efficiency, but predict that it would reduce sorting efficiency
as sorting units would have less time to interact with the material passing through them. A more
complete analysis of sorting efficiency sensitivity to changing conditions could yield
mathematical relationships for use in the model, but obtaining this data will require many tests.
Our data suggests reasonable bounds for the sorting efficiencies of equipment commonly found
in single stream recycling given observed variations in input composition. The calculated
sorting efficiencies for the magnet and eddy current were used by Ip et al. (2016) as the upper
and lower bounds for a sensitivity analysis. The effect of changing equipment parameters also
suggests ways that operators could improve the sorting efficiency of their equipment for a given
input composition. For example, increasing the eddy current rotor speed may improve grade
without compromising recovery.

61
62
Chapter 3
Optimization of MRF Design
The MRF model introduced in Chapter 2 has two functions. One evaluates how a given
input waste stream is sorted by a fixed configuration of sorting units. The second finds the best-
performing MRF configuration for a specified input composition and set of sorting equipment.
This design function uses a genetic algorithm approach to generate different configurations and
identify those with the highest overall sorting efficiency or profit. The two functions of the
model can aid MRF designers or operators in understanding plant operations, and in modifying
plant configuration, or designing new MRFs, to attain performance or profit objectives.
In this chapter, we apply the evaluate function of the model to two different MRF
configurations. We then use the design function of the model to suggest possible performance
improvements for an MRF by reorganizing the sorting units, and by adding an additional NIR
unit.

3.1 Evaluation of Single Stream MRFs

Evaluation of different MRF scenarios provides insight into how the performance of an
MRF changes under different conditions, and to what extent. This can inform plant-level
decisions such as whether to install new equipment, or where to position quality control
personnel at the outputs. A model is particularly useful for an MRF because measuring the
effects of physical changes relies on waste sampling, which, as presented in Chapter 2, is
difficult. The operators of the Spanish MRF had done some experiments with conveyor belt
speed and the angle of the ballistic separator, but changes to the physical parts of the plant take
time, and as with any factory, down time costs money, so they were interested in a way to assess
changes without having to shut down the equipment. In this section, we use the model to
evaluate two single stream MRF configurations. The first is based on the United States facilities
studied by Combs (2012), and the second is a hypothetical MRF that replaces two of the sorting

63
units in the first configuration with a ballistic separator. The scenarios presented here were
chosen to illustrate how changing sorting efficiencies and input stream composition alters the
recovery and grade of MRF outputs.

3.1.1 Modeled MRF Configurations

The first single stream MRF modeled here, Configuration A in Figure 3-1, is adapted
from the literature and represents a typical single stream sorting facility in the United States.
The sorting units and outputs are identified by name, and by the label used in the model code.
A code starting with U indicates a sorting unit, V indicates an output unit, and LI is the landfill.
This is not an exhaustive schematic of an MRF, but represents the primary mechanized sorting
steps. Other components not in this diagram that would be found at an MRF include the feeder,
which loads material onto the input conveyor at a fixed rate, a bag opener, which breaks open
plastic bags so the contents can be sorted, and balers, which compress collected material into
bundles that are bound for sale. These equipment are necessary for MRF operations but do not
separate material, so they are omitted here. Also excluded from this diagram are the people who
operate the equipment. Despite this work's focus on mechanical sorting, people played an
important role at the MRFs we studied, and were always present on the sorting floor. In addition
to operating and maintaining equipment, personnel can also hand-pick waste in either positive
(removing the desired material from the waste stream) or negative (removing contaminants
from a sorted waste stream) sorting. Negative sorting by personnel is included in Scenario 4 as
quality control (QC).

64
Figure31Single Vacuum stea M c ran A (apted foCms 2012)

PlasticFines( sent
to nfiu rion Screen 2 Mixed
elm s.t
Fgr3-,wrelLa) (U4) Paper (V3)
C tedorrugated
Cardboard

(V1) NIR (PET)

NIR (HI E

FerrousMagnet

Edd Aluminum
Current (U8) J(W7)

Resue
(sentto
landfill L1)

Figure 3-1: Single stream MRF Configuration A (adapted from Combs 2012)

At the Spanish MRF, sorting of fines, 3 -dimensional containers such as bottles and cans,

and 2-dimensional objects such as paper and cardboard is achieved using a ballistic separator,

rather than disc screens. In the second MRF configuration evaluated here, Configuration B in

Figure 3-2, we replace the scalping screen and disc screen for paper recovery with a ballistic

separator. Cardboard recovery at the Spanish MRF uses a combination of multiple types of

equipment, so we have retained the disc screen that sorts corrugated cardboard (Ul) in
configuration B. The primary role of the ballistic separator will be to separate paper, and any

cardboard not captured by the disc screen.

65
-4

<80mm

Sngle Vacuum Disc Screen 1 Ballsti 3D GissBeae es Fines (sent


Input (U)( )(U9) Screen (U3) (V2) to landfill

Plastic Film Corrugated


(VO) Cardboard
(V1) Paper and
Cardboard NIR (PET)
(0)5

% NIR (HDPE)

FerrousMagnet

Eddy Current l(nu)

Residue
(sent to
landfill Ll)

Figure 3-2: Single stream MRF Configuration B. The quality control (QC) units were modeled in
Scenario 4 only.

3.1.2 Single Stream MRF Model Scenarios

We modeled five different scenarios. Scenarios 1 and 2 use Configuration A, and


Scenarios 3, 4, and 5 use Configuration B. In scenario 4 only, quality control (QC) personnel
are added to the HDPE, PET, and Paper and Cardboard outputs. Each QC unit is modeled as
having one person who carries out negative sorting (i.e. removal of non-target material) with
an efficiency of 0.90.

Input Composition
Two input compositions from the literature were used for the scenarios. One input is
based on EPA data, and the other from samples from an MRF in Pennsylvania. Both sets of
data were discussed in Chapter 2, and presented in Table 2-5. Figure 3-3, below, compares the

66
composition of the two inputs. The EPA data was used as input in Scenarios 1, 2, 3 and 4, and
the Pennsylvania data for Scenario 5. The throughput rate for all scenarios was 24 tons per hour.

0.6

0.5

0.4
n EPA
m Pennsylvania
U 0.2

0.2

0.1

0 lt
.

Aluminum Cardboard Ferrous Glass HDPE Paper PET Plasticrilm Other

Figure 3-3: Input composition for model scenarios 1-4 (EPA) and 5 (Pennsylvania)

Sorting Efficiencies

Three sets of sorting efficiency data were used for the scenarios. Scenario 1 uses the
values in Table 3-1. These efficiencies are based on operator intuition, and represent a best-case
performance for the equipment. They also provide only the sorting efficiency for the target
material into the target output. All of the non-target materials are assumed to be perfectly sorted
into the non-target outputs.

Table 3-1: Scenario 1 target stream separation efficiencies (Combs 2012)

U0 U1 U2 U3 U4 U5 U6 U7 U8

Vacuum Disc Screen 1 Scalp Screen Glass Breaker Disc Screen 1 NIR (PET) NIR (HDPE) Magnet Eddy Current

Aluminum 0 0 0 0 0 0 0 0 0.97
Cardboard 0 0.75 0 0 0 0 0 0 0
Ferrous 0 0 0 0 0 0 0 0.98 0
Glass 0 0 0 0.97 0 0 0 0 0
HDPE 0 0 0 0 0 0 0.98 0 0
Other 0 0 0 0 0 0 0 0 0
Paper 0 0 0 0 0.94 0 0 0 0
PET 0 0 0 0 0 0.98 0 0 0
Plastic Film 0.98 0 0 0 0 0 0 0 0

67
In scenario 2, we modified the sorting efficiencies from Scenario 1 for the vacuum, disc
screens, scalping screen and glass breaker to reflect conservative estimates of non-target
material sorting efficiencies. For example, the vacuum has a target efficiency of 0.98 for plastic
film. We introduce a sorting error for both paper and cardboard in Scenario 2, assuming that
2% of incoming paper and cardboard are incorrectly removed from the waste stream by the
vacuum (see Table 3-2). Target material sorting efficiencies are highlighted for emphasis. The
efficiencies for the two NIR units, the magnet, and the eddy current are based on experimental
data from the Spanish MRF for Scenario 2.

Table 3-2: Scenarios 2 - 5 separation efficiencies (modified from Combs (2012) for UO - U4,
experimental data for U5 - U8)

UO U1 U2 U3 U4 U5 U6 U7 U8

Vacuum Disc Screen 1 Scalp Screen Glass Breaker Disc Screen 2 NIR (PET) NIR (HDPE) Magnet Eddy Current

Aluminum 0 0 0 0 0 0.03 0.03 0 0.80

Cardboard 0.02 0.75 0 0 0.02 0.05 0.05 0 0


Ferrous 0 0 0 0 0 0 0 0;92 0
Glass 0 0 0.04 0 0.08 0.08 0 0.5
HDPE 0 0 0 0 0 0.10 0 0
Paper 0.02 0.02 0 0 0.94 0.05 0.05 0 0
PET 0 0 0 0 0 0.91 0.01 0 0
Plastic Film 0 0 0 0.02 0.20 0.20 0.01 0
Other 0 0 0 0 0 0.05 0.05 0.02 0

For scenarios 3, 4, and 5, the scalping screen and disc screen 2 are replaced with a
ballistic separator, as shown in Configuration B. We modeled this unit with experimental
ballistic separator efficiencies obtained from the Spanish MRF (see Table 3-3). The sorting
efficiencies for the other equipment remain the same as in Table 3-2 for all of scenarios 3, 4,
and 5.

68
Table 3-3: Ballistic sorting efficiencies for Scenarios 3 - 5 (experimental)

Ballistic Separator Output Streams

Fines 2D 3D
Aluminum 0.03 0.04 0.93
Cardboard 0.00 0.76 0.24
Ferrous 0.03 0.04 0.93
Glass 0.51 0.12 0.36
HDPE 0.11 0.23 0.67
Other 0.50 0.25 0.25
Paper 0.00 0.76 0.24
PET 0.10 0.20 0.70
Plastic Film 0.00 0.76 0.24

3.1.3 Modeled Recovery and Grade for Scenarios

The results from the five modeled scenarios are shown in Table 3-4 and Table 3-5. For
each scenario, material recovery is given as the mass of each material type in tons per hour, and
as a fraction of the input mass of that material. A material is considered recovered if it is
collected in its target output stream. Mass is balanced in the model, so materials that are not
recovered in their target output are accounted for as contaminants in other outputs, or collected
in the landfill.

69
Table 3-4: Modeled recovery and grade for MRF scenarios 1, 2, 3 and 4

Input Scenario 1 Scenario 2


Output Output
(T/hr) Collected Recovery Grade Collected Recovery Grade
(T/hr) (T/hr)
Aluminum 0.70 0.68 0.98 1.00 0.52 0.75 0.93
Cardboard 9.55 7.16 0.75 1.00 7.02 0.73 0.98
Ferrous 0.91 0.89 0.98 1.00 0.83 0.91 1.00
Glass 3.07 2.98 0.97 1.00 2.86 0.93 1.00
HDPE 0.74 0.73 0.98 1.00 0.47 0.63 0.75
Paper 6.91 6.50 0.94 1.00 6.24 0.90 0.99
PET 0.89 0.87 0.98 1.00 0.81 0.91 0.77
Plastic Film 1.30 1.27 0.98 1.00 1.27 0.98 0.79
Other -_-
MRF Configuration A A
Input Composition EPA EPA

Target efficiencies (Combs Experimental sorting efficiencies for NIR,


Sorting Efficiencies 2012), 100% sorting efficiency magnet, and eddy current; Modified non-
for non-target materials target efficiencies for other units

Input Scenario 3 Scenario 4

Output Output
(T/hr) Collected Recovery Grade Collected Recovery Grade
(T/hr) (T/hr)
Aluminum 0.70 0.48 0.69 0.97 0.48 0.69 0.97
Cardboard 9.55 8.81 0.92 0.98 8.81 0.92 0.98
Ferrous 0.91 0.77 0.84 1.00 0.77 0.84 1.00
Glass 3.07 1.08 0.35 1.00 1.08 0.35 1.00
HDPE 0.74 0.31 0.42 0.72 0.31 0.42 0.96
Paper 6.91 5.07 0.73 0.86 5.07 0.73 0.98
PET 0.89 0.56 0.63 0.76 0.56 0.63 0.97
Plastic Film 1.30 1.27 0.98 0.79 1.27 0.98 0.79
Other - -
-

MRF Configuration B B
Input Composition EPA EPA
Experimental sorting Same as Scenario 3, but Quality Control
Sorting Efficiencies efficiencies for ballistic added to PET, HDPE and Paper outputs
separator

70
In Scenario 1, the assumption of zero incorrect sorting means that the recovery of each
material is equal to its sorting efficiency, and the grade of the outputs is 100%. In Scenario 2,
estimated incorrect sorting rates for the first four sorting units, and used experimental
efficiencies for the NIR units, magnet, and eddy current. The HDPE stream stands out for its
poor recovery and grade. The recovery of HDPE is low because 10% of HDPE is lost in the
PET stream, and then then only 70% of the remaining HDPE is recovered by NIR U6. The
grade is low because the NIR that sorts HDPE incorrectly selects for several other materials.
Most of the contamination in the HDPE stream is cardboard. This is explained by the abundance
of cardboard in the input (almost 40% by mass), and the low sorting efficiency of disc screen
1. The NIR that selects HDPE incorrectly diverts 5% of cardboard into the HDPE stream, but
due to the high amount of cardboard, this leads to 100kg of cardboard for every 470kg of HDPE
in the HDPE output.
Scenarios 3 and 4 have a ballistic separator instead of a scalping screen and disc screen
2. The sorting efficiency of the ballistic separator, determined through measurements at the
Spanish MRF, is poor (see Table 3-3). The Spanish facilities uses three connected ballistic
separators, but here we have modeled only one, assuming that the MRFs in all of the scenarios
are approximately the same size. The ballistic separator sorts based on size, and we found its
efficiencies in terms of material type by conducting a granulometric analysis of the waste, i.e.,
by measuring the size fractions of each type of material, so we could map size to type. Our
sorting efficiencies are therefore tailored to the size distribution of the materials in our samples.
In Scenario 3, the recovery of glass is very low because the glass breaker screen is downstream
from the ballistic, but the ballistic sorts 50% of incoming glass into the fines stream. In the
Spanish MRF, most of the glass entered the plant in small fragments and was collected from
the fines output from the ballistic. The use of a glass breaker screen, in contrast, assumes that
glass bottles enter a facility intact, and are only broken when they reach the glass breaker. The
improvement in grade in Scenario 4 is because of quality control personnel at the HDPE, PET,
and paper outputs. QC does not change recovery because we assume that personnel are
removing only contaminants.
The results of Scenario 5 in Table 3-5 demonstrate the impact of changing the input
composition. The PA input contains much less cardboard that then EPA input, but almost twice
as much paper, and an amount of "other" material, which could be any non-recyclable. At the

71
Spanish MRF, the other category was fine organic matter and seemingly random other
materials, such as shoes, fruit, baby diapers, and sticks. We assume that all materials in the
category other should be sorted to landfill. The high volume of incoming paper is reflected by
the low grade of the cardboard output. 2% of incoming paper is incorrectly sorted into the
cardboard by the disc screen, but since the incoming amount of cardboard is very low, the
cardboard output is almost 50% paper.

Table 3-5: Modeled recovery and grade of materials for Scenario 5

Input Scenario 5

T/hr T/hr Recovery Grade

Aluminum 0.51 0.35 0.69 0.94


Cardboard 0.32 0.24 0.74 0.48
Ferrous 1.30 1.10 0.85 1.00
Glass 4.95 1.75 0.35 1.00
HDPE 1.60 0.68 0.43 0.88
Paper 13.33 9.78 0.73 0.89
PET 1.16 0.73 0.63 0.77
Plastic Film - - -
-

Other 0.86 0.64 0.74 0.09


MRF Configuration B

Input Composition Pennsylvania

Sorting Efficiencies Same as Scenario 3, no Quality


Control

An MRF operator could use modeled scenarios such as those presented above to prepare
for input changes, or to adapt to new technology. Around Christmas, for example, the amount
of paper in the waste stream increases significantly from discarded wrapping paper. Scenario 5
shows that the MRF with Configuration B is not well equipped to sort paper, as almost 30% of
it goes to landfill because it is diverted into the 3D output by the ballistic separator. A secondary
sorting step could be installed downstream from the ballistic to capture more paper, or the angle
of the ballistic separator could be adjusted to recover more paper.

72
3.2 MRF Performance Metrics

In addition to the recovery and grade of outputs, the model calculates overall performance
metrics for a scenario. These metrics are profit and overall sorting efficiency. They provide a
basis of comparison between MRF scenarios, and for finding the best configuration using the
optimization function of the model.

3.2.1 Calculation of Profit

The costs that can be included in the profit calculation include the purchase price, energy
use, and maintenance costs of equipment, wages of plant personnel, and the cost of sending
residual material to landfill. Revenues include tipping fees charged to communities by the MRF,
sales of recovered material, and subsidies for recovered material if applicable. For this analysis,
we used a simplified profit calculation that includes only the revenues from sale of materials,
and the cost of sending residues to landfill. The calculated profits presented here are therefore
not representative of actual plant profits, but may be used to compare different configurations
assuming that other factors remain the same.
The revenues associated with the sale of each material is a user-defined model input to
allow for different materials markets. In Spain, as in other European Union countries, there are
two sources of revenue for recovered materials; the market price, and a subsidy based on the
recovery of materials. This subsidy is paid by an intermediate organization that helps packaging
manufacturers fulfill their legal obligations under the principle of extended producer
responsibility (European Comission 2008). As this analysis is focused on MRFs operating in
the United States, where there is no national recycling policy, we used only the market price.
The selling prices of materials for this analysis are based on information from an MRF
in Massachusetts (see Table 3-6). Both glass and landfill residue are a cost to this facility. Glass
bottles that enter the single stream recycling process are often smashed in transport or the first
phases of separation, and the resulting small pieces of glass are very hard to separate from other
small debris such as bottle caps and pieces of aluminum foil. The quality of the glass output
from a single stream MRF is therefore too low to be remade into glass bottles, and is instead
used as low-value inert fill in roadways or similar applications. The value of this recovered

73
glass is so low in the local market that the Massachusetts MRF has to pay a company to haul it
away. The cost of landfill is based on the fee that the MRF has to pay to dispose of the residual
material that cannot be recycled.

Table 3-6: Selling price of recovered materials in dollars per metric ton (source: MRF in MA, 2016).

Materials $/Ton
Cardboard $127
Newspaper $88
Mixed Paper $77
PET $204
HDPE Natural $573
HDPE Colored $320
Mixed Plastic $198
Aluminum $1,466
Other Metals $143
Glass ($40)
Residue ($83)

The costs in Table 3-6 are used as the prices in the model, with the following exceptions:
Glass is modeled as having zero value, rather than a negative price. Ferrous is modeled as
$143/ton. All paper is modeled as $77/ton, as our input categories do not distinguish between
types of paper. The price of HDPE is $447/ton, the average of the colored and natural HDPE
prices in the table, again because we do not have only aggregate HDPE composition data.
Finally, plastic film, which includes plastic shopping bags, is modeled as having zero selling
price. We assume it is collected to protect the other equipment in the plant.

Potential Revenues in Input

The combination of price per ton and abundance in the waste stream presents a challenge
for waste managers. Figure 3-1 shows the composition of input to the Massachusetts facility as
the fraction of each material by mass. The secondary axis plots the potential value of each
material in one ton of total input, given the prices in Table 3-6. For example, PET is 2.3% of
the input material, and sells for $204/Ton. One ton of mixed input will therefore contain 204kg
of PET, which, if all of it is recovered, has a value of $4.61. The difficulty that this data
highlights is how an MRF operator should prioritize sorting materials from an economic

74
perspective. Paper and cardboard are important sources of revenue due to their volume, but
plastic and metal containers, present in much smaller quantities, are still important to recover
due to their high value.

35% $30

0
- $25
30%
$20
25%
0 . $15

20% - - $10
E 0 I0
o 5 - - $5 g
C 15% -a

10% -
-

I -$5
10% --
$(

)
00

0% $(15)

U Input Composition OValue of Material in 1 Ton of Input

Figure 3-4: Composition and value of materials in single stream MRF input (source: MRF in
Massachusetts, 2016)

Limits to Recovering Potential Revenue

If it were possible to recover 100% of incoming materials into their target outputs, then
the value of the incoming material would equal the revenues. In reality, however, material
recovery is compromised by imperfect sorting of target materials into outputs, and by upstream
losses. There may also be tradeoffs in the recovery of one material over another. For example,
one of the three outputs of the ballistic separator is the small particles that fall through the screen
and are sent to landfill. At the Spanish MRF, the size of the fines screens was 80mm, which is
larger than the diameter of most soda cans. The purpose of the fines screen is to remove organic
matter and other non-recyclables from the waste stream so they do not contaminate downstream
recovered outputs, however some valuable materials will also fall through the screen. In fact,
the Spanish MRF recently installed an extra eddy current separator on their fines stream to

75
recover aluminum. Reducing the screen size would reduce the number of valuable materials
lost, but would also remove unwanted material less efficiently.

Grade Dependence of Market Price

The grade of recovered materials is a factor in both the downstream use of the material,
and the selling price. Although the goal of separation systems is to separate out pure materials,
which could be valued the same as virgin materials, we have shown that real separation systems
produce outputs that have some contamination. Sorted materials have to meet standards set by
industry groups or individual buyers. For example, the Institute for Scrap Recycling Industries
(ISRI) sets standards for scrap material in the United States. The Spanish MRF has to meet
output requirements set by Ecoembes, a recycling organization that helps companies meet their
obligations under EU waste laws. The Ecoembes output requirements are provided as reference
in the Appendix.
One approach to modeling the price of materials based on standards is using a stepped
system, where materials that meet a certain purity tier are valued at a fixed price. Wolf (2011),
suggests that for materials with a large selection of purity points, a linear relationship between
price and purity may be established. This linear price model may be applicable for a large
market with buyers of many different qualities of scrap, but for a smaller market with a limited
number of buyers, the stepped pricing model is more appropriate. Through our work in Spain,
we learned that outputs that did not meet requirements were still sold, rather than being sent to
landfill, but we did not obtain price data.
The selling price of materials is modeled as linearly dependent on grade between a lower
grade sale threshold, and the purity requirement. We set the sale threshold at 0.5, as in Wolf
(2011). The required grade of materials is set by the model user. The price of material m in
output stream i is modeled as:

Pi(m) f(Gm)P,(,ina1 (3-1)

where f(Gm) = 1 if the concentration requirements are met. If not, f(Gm) < 1 is a penalty
on the selling price as a function of grade (Ip et al. 2016):

76
F- - 4

2Gm-Gm

f(Gm) = G 0, Gm > 0.5 (3-2)


m;05

where G4ris the required grade of material m. Using a purity threshold of 0.75, the selling price
for different grades of cardboard, PET, HDPE, and aluminum is shown in Figure 3-5.

$1,600

$1,400

$1,200

' $1,000

- $800

. $600

$400

$200

$0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Grade

-a-Cardboard -B-PET -.- HDPE -4-Aluminum

Figure 3-5: Modeled selling price of materials as a function of grade with a purity threshold of 0.75

For the optimizations in this section, we used a purity threshold of 0.75 for most
materials, which is below what would normally be acceptable for an output. Based on our
experience at MRFs in Spain and the United States, most MRFs rely on quality control
personnel to achieve the required grade for their outputs. For example, in the light packaging
recovery section of the Spanish MRF, personnel were placed on the PET, HDPE, and aluminum
output lines to improve the grade. Using hand-sorting is very flexible, as people can be
reassigned to different outputs to compensate for problems in the plant, or to respond quickly
to changing output requirements from different buyers. The objective for this analysis was to

77
model the performance of automated sorting equipment only, so we set the purity threshold to
what was achievable given the experimental sorting efficiencies. In an real MRF, the quality
thresholds would be higher, and QC personnel would be used to achieve those thresholds. We
did not set a purity threshold for glass, as we assume the recovered glass will be used as fill for
roads or as landfill cover, so would not have a strict composition requirement.

3.2.2 Calculation of Overall Sorting Efficiency

The overall efficiency of an MRF is a single metric that measures the ability of the
facility to recover different materials. The overall efficiency, E, of a modeled MRF is:

rniMm (3-3)

where ym is the quantity of external input of material m entering the system, and fwn) is the
amount that is correctly sorted (Ip et al. 2016).

3.2.3 Performance of MRF Scenarios

The metrics of efficiency and profit for the five modeled scenarios show that the ballistic
separator causes a decrease in sorting efficiency for scenarios 3-5 (see Table 3-7). The reduced
volume of cardboard in the Pennsylvania input led to a lower profit, as cardboard is more
valuable than paper. The potential revenue contained in the EPA input (Scenarios 1-4) was
$3409/hour, and the potential revenue in the PA input (Scenario 5) was $2948/hour. Both
Scenarios 3 and 5, which were the same except for the input streams, captured about 73% of
the potential revenue, despite a more than 10% difference in sorting efficiency. The lower
sorting efficiency in Scenario 5 is due in part to paper, which was abundant but only recovered
at 73%. Scenario 5 also lacked the plastic film, which had no value but was sorted efficiently
in Scenario 3, and instead contained the other material category, which also had zero value, but
a low recovery.

78
Table 3-7: Overall sorting efficiency, revenues, and costs for Scenarios 1-5

Scenario Overall Efficiency Revenue Landfill Cost Profit ($/hr)


1 87.6% $2531 $249 $2,282
2 83.2% $2,337 $258 $2,079
3 76.3% $2,488 $341 $2,147
4 76.3% $2,545 $424 $2,121
5 63.8% $2,148 $568 $1,580

In addition to reflecting the impact of changing input composition, the data in Table 3-7
show the financial tradeoff between recovery and grade. In scenarios 3 and 4, an MRF with a
ballistic separator (Configuration B in Figure 3-2) is evaluated first with purely mechanical
separation in Scenario 3, and then with one quality control person at each of the HDPE, PET,
and Paper and Cardboard outputs in Scenario 4. The results show that the addition of quality
control personnel greatly improves the purity of the target outputs in Scenario 4, but QC has
the opposite effect on profit. This is explained by the grade threshold. The HDPE output in
Scenario 3 did not meet the 75% grade requirement. Using the equation for price from section
3.2.1, the price of the below-spec HDPE is modeled as:

PHDPE - ($) (2(0.72) 0.7) = $412/Ton

This calculation assumes that the MRF has access to a market where buyers exist for a
range of output qualities, and where, as for scrap metal, the price of plastics is approximately
linear with quality.
Though the quality of the PET in Scenario 3 is low, it meets the 75% threshold, so, for
the purposes of this analysis, sells for its nominal price of $204/ton. When the QC is added in
Scenario 4, the selling price of the HDPE increases to $447/ton, but the total mass of the outputs,
including impurities, is reduced. If the pre-QC stream already meets the concentration
requirements, as it did for PET and Paper, then removal of the contaminants has the unintended
effect of reducing the revenue from that output. Setting a fixed price above a purity threshold
therefore provides a disincentive to achieve grades better than the threshold.

79
3.2 Optimization of MRF Performance

The design function of the MRF model uses a genetic algorithm approach to evaluate
the performance of possible configurations for a set of sorting equipment and a given input, and
outputs the best-performing configurations (Testa 2015). Rather than defining an MRF
configuration, the user defines only the set of sorting, input, and output units, and the model
finds the best way to connect them. The set of units used in this thesis is shown in Table 3-8,
with the ID used in the model. The configurations of different scenarios include a subset of
these units, and the sorting efficiencies for the units, as well as the composition of the input, are
specified for each scenario. NIR units can be programmed to target a range of plastics and
paper. Two of the NIR units, U5 and U6, are assumed to be programmed to select for PET and
HDPE, respectively. This means that they have an assigned output unit as one of the outputs
from the sorting step. The third NIR, U10, has no designated output. In the optimization, the
model determines which materials to route into the two output streams of NIR U10.

Table 3-8: Sorting, output, and input units for model scenarios

Sorting Units Output Units

No. of
ID Equipment Type Output Target Material(s) ID Material Collected
Streams
U0 Vacuum 2 Plastic Film/Bags VO Plastic Film/Bags
Ul Disc Screen 2 Cardboard (OCC) V1 Cardboard (OCC)
U2 Scalping Screen 2 Fines V2 Glass (mixed colors)
U3 Glass Breaker 2 Glass V3 Paper and Cardboard
U4 Disc Screen 2 Paper V4 PET
U5 NIR (PET) 2 PET V5 HDPE
U6 NIR (HDPE) 2 HDPE V6 Ferrous
U7 Magnet 2 Ferrous Metals V7 Aluminum
U8 Eddy Current 2 Aluminum Li Landfill
U9 Ballistic Separator 3 2D & 3D Materials Input Unit
U10 NIR 2 Undefined EG Input to MRF

80
The user input to the optimization may be visualized as a set of sorting units with
dangling connections, illustrated in Figure 3-6. The model calculates the best way to connect
the sorting units to each other to maximize either profit, overall sorting efficiency, or a weighted
combination of both. The user can place restrictions, in the form of precedence constraints, on
possible combinations to account for real operating limitations. In this analysis, we set two
precedence constraints. The first is that the vacuum (UO) must be the first sorting unit of any
configuration, so the input stream, EO, is connected to UO. This constraint is shown as a dashed
arrow between EO and UO in Figure 3-6. The vacuum removes plastic film, which includes
plastic bags and other light, flexible, low density polyethylene (LDPE) products. Plastic bags
are one of the greatest challenges for MRFs, as they become entangled in rotating shafts and
other moving parts of sorting equipment, so they need to be removed from the input stream
immediately. The Spanish MRF employs people to stand at the input stream and manually
remove large bags from the waste stream. In the MRF modeled here, a large hanging vacuum
is positioned over the input conveyor to remove films.
The second precedence constraint is that the magnet (U7) must always precede the eddy
current (U8), though they do not need to be directly connected. The presence of ferrous material
can disrupt the magnetic field in the eddy current, reducing its sorting efficiency of aluminum.
The magnetic force on the ferrous materials in the eddy current can also cause them to become
stuck to the conveyor near the magnetic rotor. The induced electric current in the ferrous metals
causes them to become very hot, which can melt the conveyor belt and damage the equipment.

Ul U6U9
U8

UOV1 - -V5 L1 V3 V70

VO U7 - +U 1

V6 V4 V2

Figure 3-6: Unordered units for use in the optimization function of the MRF model

81
Ten dangling connections yield 10! = 3.6 million possible configurations for the sorting
equipment in Figure 3-6. The model uses a genetic algorithm approach to quickly find the best
configuration. Rather than generating all possible configurations at once, the genetic algorithm
starts with a randomly generated initial population, PI, of N individuals, from which it generates
a related offspring population, Qt. The model then evaluates this offspring population for
performance and efficiency, selects the N best performing individuals and copies them to
population Pt+1 . Performance is measured as a weighted combination of profit and overall
sorting efficiency. This process is repeated until termination when t = Tmax (Testa 2015). For
this analysis we used population size P = 500, and Tnax = 100 iterations.
In this section, we use the design function of the model to find the best configuration for
the MRF evaluated above in Scenario 3, and compare the optimized and original performance
metrics. We then explore possible improvements to the performance of the MRF by adding a
third NIR unit that enables a recirculation loop.

3.2.4 Modeled Best MRF Design

The modeled best configurations for the Scenario 3 MRF are shown in Figure 3-7, as
well as the original Scenario 3 configuration. For Optimization A, the relative weight of profit
in the performance metric was set to one, and in Optimization B, the weight of sorting efficiency
was set to one. These two conditions represent the maximum feasible profit and sorting
efficiency for these model inputs and constraints.

82
EO U0 U3 U7 U1 us U9 U6 us Ll

A) Profit

V0 V2 V6 -V..Ll V5 V4

ED U0 U3 U7 us U9 U1 U6 us Ll

B) Efficiency
V0 V2 V6 Ll V5 V4

EO U0 U1 U9U3 us U6 U7 us Ll

Scenario 3 MRF Design

V0 V1 Ll V2 V4 V5 V6

Figure 3-7: Optimized configurations for Scenario 3 sorting equipment and input. Performance metric
weighted by profit =1 (Optimization A) and sorting efficiency = 1 (Optimization B), with comparison
with original, non-optimal MRF design

3.2.5 Modeled Best MRF with Recirculation

An MRF operator might want to know if an additional sorting unit would be worth the
investment and operating cost, and where to place the unit in an existing system. In this
example, we add an NIR unit to the set of sorting equipment, but do not define the target outputs.
The unit (U 10) therefore has two dangling connections. As the other outputs units are, in this
case, assigned to existing sorting units, the additional NIR adds a recirculation loop in which
valuable materials are directed back into an upstream sorting unit to run through the system
again.

83
EO U0 U3 U1 U7 us U9 U6 us U10 LI

Q Profit .0VV2V6 L1 V V5 V4

ED U0 U3 U7 us U9 U1 U6 us U10 LI

D) Efficiency
VC V2 V6 L1 V V1 V5 V4

Figure 3-8: Optimized MRF designs for MRF in Scenario 3 with the addition of an NIR unit (U 10) with
no specified target material. Designs were optimized for profit (Optimization C) and efficiency
(Optimization D).

3.2.6 Performance of Optimized Configurations

The performance of the four MRF configurations shown in the figures above is
summarized in Table 3-9. These results show a dramatic improvement over the original MRF
in Scenario 3 (see Table 3-7), for which the sorting efficiency was 76.3%, and the profit $2,147.
Compared with the original design, the optimized configurations reduce the mass of materials
sent to landfill. Some of this improvement comes from the recovery of glass, of which only
1.08 T/hr is collected in Scenario 3, but 2.98 T/hr is collected in the optimized MRFs. As glass
is modeled with a selling price of zero, this does not impact the revenue from sale of materials,
but reduces the cost of landfill and boosts overall sorting efficiency. The optimized MRFs also
improves the machine-sorted quality of the HDPE to above the 75% specification level, and
increases the mass (though decreases the quality from 97% to 92%) of the aluminum output,
leading to higher revenues from the sale of aluminum.
A second difference between the optimized configurations and the original Scenario 3
design is the placement of the NIR units relative to the magnet and eddy current. The NIR units
have higher rates of incorrect sorting, so placing them last reduces the losses of ferrous and
aluminum. This finding agrees with the best design for the LPRS in the Spanish MRF, included
in the Appendix.

84
Table 3-9: Performance metrics and output stream data for modeled best MRF designs in Optimizations
A - D.

MRF Configuration Optimization A Optimization B

Priority Metric Profit Efficiency

Efficiency 84.9% 85.3%

Profit ($/hr) $2,599 $2,535

Output Output
Revenue Revenue
Output Unit Collected Grade Collected Grade
($/hr) ($/hr)
(T/hr) (T/hr)
I

VO Plastic Film 1.60 79% $0 1.60 79% $0


V1 Cardboard 7.16 98% $909 1.69 98% $214

V2 Glass 2.98 100% $0 2.98 100% $0


V3 Paper/Cardboard 7.25 95% $558 12.71 97% $979

V4 PET 0.68 83% $138 0.68 83% $138

V5 HDPE 0.46 76% $205 0.46 76% $205

V6 Ferrous 0.84 100% $120 0.84 I 100% $120

V7 Aluminum 0.60 92% $879 0.60 92% $879

Li Landfill 2.52 NA -$209 2.52 NA -$209

MRF Configuration Optimization C Optimization D


Priority Metric Profit Efficiency
Efficiency 93.3% 93.9%
Profit ($/hr) $3,179 $2,879
Output Output
Grade Revenue Collected Grade Revenue
Output Unit Collected
(T/hr) ($hr) (T/hr) ($/hr)

VO Plastic Film 1.60 79% $0 1.60 79% $0

V1 Cardboard 7.57 98% $962 1.79 98% $227

V2 Glass 2.98 100% $0 2.99 100% $0

V3 Paper/Cardboard 8.67 95% $668 14.45 97% $1,113

V4 PET 0.73 81% $150 0.73 81% $149

V5 HDPE 0.54 76% $240 0.54 76% $240

V6 Ferrous 0.91 100% $130 0.91 100% $130


I
V7 Aluminum 0.72 93% $1,060 0.72 94% $1,049

Li Landfill 0.36 NA -$29 0.35 NA -$29

85
The difference in performance between Optimizations A and B is how paper and
cardboard are sorted. The cardboard selected by the disc screen U1 is modeled as old corrugated
cardboard (OCC) and priced at $127/ton. The separation efficiency of Ul is moderate, selecting
75% of the cardboard that enters, but the rate of incorrect sorting is low, with only 2% of paper
incorrectly sorted into the cardboard output. The ballistic separator, U9, recovers 76% of both
paper and cardboard, which when combined are priced as mixed paper at $77/ton. Placing the
disc screen upstream of the ballistic results in more cardboard recovered into the valuable OCC
stream, but there a small penalty in the sorting efficiency of paper. Placing the ballistic upstream
of the disc screen results in more cardboard being sorted into the mixed paper and cardboard
output (V3), which sells for a lower price than the pure OCC, but slightly improves the overall
sorting efficiency as both cardboard and paper are target materials for that output. We note that
this result is partly an artifact of how we modeled the outputs, with cardboard as a target
material for both, but more valuable when sorted into a separate stream.
Optimizations C and D, with a recirculation loop, show a similar relationship with the
higher profit based on separating the cardboard out first, whereas the overall sorting efficiency
improves if the ballistic separator is earlier in the sorting sequence to avoid loss of paper. The
primary benefit of the additional NIR unit is the improved recovery of valuable materials due
to the recirculation loop. This both improves the overall sorting efficiency of the MRF, and
reduces the cost of landfill.

3.3 Design Principles for MRFs

The modeled scenarios suggest some design rules for MRFs to maximize sorting efficiency
or profit. The priority of these rules will depend on how materials are valued and what levels
of contamination are acceptable, conditions that could be evaluated using the model. Our
proposed rules are the following:
1. Separate out any non-valuable materials early in the sequence to prevent them from
contaminating downstream outputs. Common contaminants are fine organic matter and
other small particles, plastic bags and film, and broken glass.

86
2. Position the units that sort materials entering the waste stream in large amounts near the
start of the sorting sequence. Overall sorting efficiency is based on the total mass of
materials recovered, so maximizing recovery of the most abundant materials yields
higher overall efficiency, and prevents contamination of downstream outputs.
3. Position units with high non-target efficiencies towards the end of the sequence to avoid
loss of other valuable materials.
4. Use recirculation loops to improve recovery, and route those loops from the last sorting
unit in the sequence.

87
88
Chapter 4
Application of MRF Model Results to
Recycling Life Cycle Assessment
The MRF model in this paper has valuable applications for the managers and designers
of sorting facilities. From a broader waste management perspective, more complete information
about the recovery and grade of materials sorted by an MRF could also contribute to better and
more sustainable end-of-life materials decisions. Waste managers in states and communities in
the United States need to weigh financial and regulatory constraints against environmental
benefits and public opinion as they decide what types of recycling programs to implement. The
EPA developed a waste life cycle assessment (LCA) tool to help these managers quantify the
greenhouse gas (GHG) emission impacts of their waste management plans. The Waste
Reduction Model (WARM) relies on estimates of material recovery to calculate the impact of
recycling compared to landfill or incineration. Our MRF model could be used to tailor those
recovery estimates to a particular location, sorting facility, or waste stream, and thereby
improve the accuracy of LCA results.
This chapter introduces recycling LCA, and some of the uncertainties associated with
it. We then compare the material recovery values used in WARM with modeled and
experimental results for sorting in an MRF, and examine how assumptions about the grade of
recovered materials determine the way that recycling is modeled. The chapter concludes with a
discussion of the benefits and limitations of using the MRF model to inform recycling LCA,
and implications for recycling policy in the United States.

4.1 Life Cycle Assessment for Recycling

Life cycle assessment accounts for the energy and emissions associated with each step
in a product's life cycle. In LCA for waste management, including recycling, the life cycle

89
begins when an end-of-life product enters the waste stream. The life cycle of a recycled material
ends when the recyclate is used to manufacture a new product. Figure 1-3 in Chapter 1 illustrates
the flow of materials through each step in the life cycle, including material losses. Calculation
of the total energy and emissions associated with a recycled material must account for the
following:
1. Process and Transportation: Each step, and the transportation between steps, has an
associated energy and emissions cost. The magnitude of these costs will depend on the
processes themselves, the physical distance between steps, the type of transportation
used to move materials, and the mass of the material being recycled. Some processing
steps also have direct, non-energy emissions, such as the perfluorocarbons generated
during aluminum smelting (U.S. EPA 2016c).
2. Material losses: There are several places where materials collected for recycling may
be instead diverted to landfill. These include residues from sorting and reprocessing,
and yield losses from material production, fabrication, and product manufacturing.
Finding the total energy and emissions in terms of end-of-life products, requires
accounting for the materials lost at each step.
3. Closed or Open Loop Recycling: There are two possible fates for recyclate, the
material retained through the collection, sorting and reprocessing steps (streams (g) and
(k) in Figure 1-3). In closed loop recycling, reprocessed materials are used to substitute
for virgin materials in the manufacture of the original product. The end-of-life recycling
rate, EOL-RR, quantifies the fraction of end-of-life materials that are used in closed
loop recycling. In open loop recycling, reprocessed materials are used to substitute for
different virgin materials in a secondary product. The recycling rate for open-loop
recycling is indicated as EOL-RRo (Edwards and Schelling 1996; Graedel 2011).

As an example, the EPA's calculations for total emissions from the manufacture of
virgin and recycled steel and aluminum cans are shown in Table 4-1. Based on these values,
the difference in emissions between producing 1 ton of new cans from recycled materials and
1 ton of new cans from virgin material is a savings of 10 MTCO 2E for aluminum, and 2
MTCO 2E for steel. These calculations assume closed loop recycling, where recycled cans are

90
melted into ingots that can substitute directly for virgin material in the manufacture of new
cans.

Table 4-1: Emissions from virgin and recycled aluminum and steel can production (U.S. EPA 2016d)

Product Manufacture Product Manufacture Difference Between


Using 100% Virgin Using 100% Recycled Recycled and Virgin
Inputs Inputs Manufacture
(MTCO 2E/Ton) (MTCO 2E/Ton) (MTCO 2E/Ton)

Transport Prcess Process Transport Pocess Process Transport Pocess


Process
Energy Energy Ene Energy Energy Eng Energy Energy Energy
Energy Energy Eeg

Alumipum Cans 8.01 0.10 4.10 2.12 0.07 0.00 -5.90 -0.03 -4.10
Steel Cans 2.68 0.41 0.96 0.68 0.35 0.96 -2.00 -0.06 0.00

The values in the last 3 columns of Table 4-1 are an upper bound for the energy benefit
from recycling, as they do not account for the losses between end-of-life and input to the
manufacturing process. To put the emissions benefits in terms of material that enters the waste
stream, a recycling rate is applied that reflects the fraction of end-of-life material that becomes
recyclate through collection, sorting, and reprocessing. In this case, the reprocessing step would
be melting recovered cans back into aluminum or steel ingots. The total retention rate is the
end-of-life recycling rate, EOL-RR. For the case of aluminum cans, the EPA assumes a 100%
collection and sorting recovery rate, and a reprocessing retention rate of 93%, so the end-of-life
recycling rate for aluminum in this example is the product of the two, 93%. For steel cans, the
EOL-RR is 98%.
The EPA defines an emission factor for recycling as the GHG benefit from using
recycled rather than virgin inputs per ton of end-of-life material. It is calculated by multiplying
the difference between recycled and virgin manufacture by the end-of-life recycling rate of the
material. In this example, the emissions factor for aluminum cans is
(-10 ton recyclate) \
MTCO 2 E .93 ton ton rec
end of life cans I
MTCO E 2
ton end of life cans
, and for steel cans it is

-1.96 MTCO 2E/ton. The negative sign indicates an emissions savings from recycling. For
comparison, net emissions from sending 1 ton of end-of-life aluminum or steel cans to landfill
is estimated to be 0.02 MTCO 2E. This estimate is based only on the emissions from
transportation to landfill, as metals do not generate methane when landfilled (U.S. EPA 2016d).

91
4.1.1 Uncertainty in Waste Management LCA

The calculations above appear straightforward, but they are sensitive to the assumptions
embedded in the life cycle assessment. For metals with high embodied energies, particularly
aluminum, LCA studies show that recycling consistently has a lower impact than incineration
or landfill. For paper, cardboard, and plastics, however, the findings are more nuanced. These
materials have a high carbon content, making them good fuel for incinerators. Also, they are
often recycled in open loop systems, so rather than replacing virgin materials in the original
product, they may substitute for different materials that have a lower embodied energy, and
therefore realize a smaller savings, or none at all. The benefits of recycling these materials
relative to landfill and incineration are more sensitive to boundaries and assumptions, and the
literature reflects this.
Bjbrklund & Finnveden identify factors that explain why, in some cases, incineration
and landfill have lower net emissions than recycling. One is the energy sources avoided by
energy recovery from incineration. The GHG savings associated with incineration of paper and
cardboard, for example, may be greater than recycling if energy from a high-carbon fuel source
such as coal is displaced. Similarly, plastics yield benefits if incinerated when energy from coal
is displaced, but should be recycled if natural gas is the power generation source (Rigamonti et
al. 2014). The efficiency of the waste-to-energy plant is also important in this calculation. A
second factor is the time perspective assumed for landfills. If a short time horizon is considered,
then only biodegradable materials will decompose in landfill and emit methane. If the time
horizon is lengthened to hundreds or thousands of years, however, plastics will also break down
and emit carbon into the atmosphere (Bj6rklund and Finnveden 2005). A third factor that
impacts the management of renewable materials, mostly paper and cardboard, is forest carbon
storage, or the amount of carbon stored by trees if they are not harvested to make paper. This
can be counted as an additional benefit of recycling paper and cardboard.
Finding the energy and GHG impacts of landfilling or incinerating waste is a complex
calculation, and the purpose of this introduction is simply to highlight that LCA results for those
management options are sensitive to a range of assumptions. Any application of LCA results
for decision making about waste management needs to involve an understanding of the
methodology and limitations of the assessment. The focus of this paper, however, is on
recycling, which has its own set uncertainties and assumptions in life cycle analysis.

92
4.1.2 Assumptions in Recycling LCA

There are several points in the life cycle of a recycled material where LCA relies on
assumptions that may not always be valid. In some cases, a simplifying assumption is used due
to lack of resources or data, for example in assuming closed or open loop recycling for a product
that in reality could follow either path. In others, however, the underlying phenomenon is not
well understood, and is the subject of ongoing research. An example of this is the displacement
rate of raw materials due to recycled inputs. This section introduces some sources of uncertainty
in recycling LCA.

Substitution Ratio

The substitution ratio determines the amount of primary material production that is
avoided from recycling an amount of end-of-life material. It is the basis for the benefits from
recycling, and is the product of three factors. The first is the end-of-life recycling rate (EOL-
RR), or the fraction of end-of-life product that enters the manufacturing process. The EOL-RR
is a function of collection, sorting, and reprocessing retention rates, and is partly determined by
material recovery in MRFs. The second is the material quality loss, which determines the
material properties of the recyclate. This in turn determines the amount of recyclate that can be
incorporated into a new product, and whether the recyclate is used in open or closed loop
recycling. The third factor is the market substitution rate, or the amount of primary production
that is displaced as a result of using recycled inputs. The market substitution rate reflects market
elasticity. It is almost always assumed that the presence of recycled products does not affect the
market for products, so that this ratio is 1:1, but this may not be a valid assumption (Turner,
Williams, and Kemp 2015).

Recycled Content

Most material production includes some recycled content. The average recycled content
of some common metals in Table 4-2, and the EPA assumes a recycled content of 9.8% for
corrugated cardboard, 5% for glass, 10% for HDPE and 3% for PET (U.S. EPA 2016d). The
"virgin" material used in manufacturing products is therefore seldom 100% virgin material, but
rather the mix of recycled and virgin content in the supply. The market has an impact on

93
recycled content, as if there is a growing demand for a product, then the amount of available
scrap from EOL products may be inadequate to substitute for virgin materials (Graedel 2011).

Table 4-2: Global average recycled content of metals (Graedel 2011)

Average Recycled
Content
Magnesium 25-50

%
Aluminum 25-50
Titanium 10-25%
%
Iron 25-50
%

Nickel 25-50
%

Copper 10-25
%

Zinc 10-25%

Primary Material Displacement

A significant assumption that underlies most recycling LCAs is that the amount of
recycled material used in the manufacture of a product displaces an equal amount of virgin
material production. This one-to-one displacement relies on a market where the presence of
recycled material in products does not affect the demand. In other words, there is a fixed amount
of demand for a product, so the demand met with recycled goods displaces the need for virgin
materials. Figure 4-1 shows a simple model for how recycling displaces primary materials. The
dashed line indicates primary material production that does not occur because the new material
is made from recycled input. Despite the frequency with which this assumption is made, a
market analysis shows that displacement may in reality be small, or even negative if sales of
recycled goods lead to additional sales of new products (Zink, Geyer, and Startz 2015). In the
case where the displacement is negative, recycling effectively leads to increased material use,
and a subsequent increase in emissions and energy costs.

94
Primary Production Product Lan dfiI

Collection and

--.----.----------.- eroesn Material

&
L -------------------------------------- - product
Production
-------------------- markets

Figure 4-1: Displaced production of a recycled product (Zink, Geyer, and Startz 2015)

Closed and Open Loop Recycling

In closed-loop recycling, recycled materials are manufactured back into the original
product and replace virgin material of the same type. An example of closed-loop recycling is
when aluminum recovered from aluminum cans is used to manufacture new aluminum cans. In
open-loop recycling, in contrast, the products made with recycled material are different from
the original products. Recycled material is used to manufacture a secondary product, and
displaces other types of virgin material that may have lower embodied energies than the primary
material. An example of open-loop recycling is the use of recycled plastic to replace wood.
The benefits from open-loop recycling are generally smaller than closed-loop recycling.
In one analysis, plastic from domestic appliances are recovered at 90% (i.e., EOL-RRpastic

-
0.9), and used in a closed loop to substitute for virgin plastic in new appliances. It is assumed
that the plastic recyclate can substitute for the same mass of virgin plastic, and that the new
plastic parts are made from 100% recycled plastic. The substitution ratio is therefore 1 kg of
end-of-life plastic to 0.9kg of displaced virgin plastic. In an open loop case, the recovered
plastic, a mix of PET and PP, replaces virgin cast iron and wood to make a park bench.
Household plastics are recovered at 80%, and every 0.8kg of recovered plastics substitutes for
0.3kg of wood and 0.2kg cast iron in the manufacture of a park bench. More plastic recyclate
is needed than if using wood and cast iron because the recycled plastic is not as strong as the
original materials. The substitution ratio for this process is then 1:0.5. The calculated energy
saved by recycling 1kg of waste plastic in the closed-loop scenario is about 7 times greater than
recycling 1kg of plastic in the open loop scenario (Huysman et al. 2015).
The quality of recovered material can determine whether closed or open loop recycling
is appropriate, and sorting at an MRF may not provide the required high level of purity. For
example, less than half of the glass recovered from single-stream MRFs is recycled into new

95
glass. It is instead used for applications such as landfill cover or construction fill (Cimpan,
Maul, et al. 2015). In the scenarios modeled in this paper, glass is sorted by a glass breaker
screen with a high sorting efficiency and no incorrect sorting, so the modeled grade is 100%.
At the Spanish MRF, in contrast, most of the glass than enters the plant is smashed by the
processing equipment, and the resulting small pieces are very difficult to separate from small
particles of organic matter, bottle caps, and other small pieces of debris. The resulting mixture
of glass and non-glass contaminants is suitable only for low-quality applications.
Paper and plastic are also hard to separate at qualities required for closed-loop recycling.
Paper poses a challenge as high recovery rates are linked to lower-quality outputs (Miranda,
Monte, and Blanco 2013). This may be explained by a tradeoff between recovery and grade
based on material source. If only office paper is collected for recycling, then the quality of the
recovered paper is high, but the overall recovery rate of paper is low. If, as in the EU, high
recovery rates are mandated for materials, then a broader range of paper types, including
newsprint and paper from household waste, must be recovered. This degrades the overall
quality of the recycled paper stream.
The comingling of recyclables has a further detrimental effect on the quality of paper,
as otherwise clean paper may be contaminated by liquids or foods from other containers. For
example, a European association of paper mills limits the unusable content of incoming
recovered paper to between 1% and 6%, but one study found that the average unusable content
for paper recovered from comingled streams was about 8% (Miranda, Monte, and Blanco 2013).
At the Spanish MRF, the quality requirement for paper was 97%. Sorting with an NIR yielded
94% purity, so quality control personnel were used to pick impurities out of the paper stream.
The Massachusetts MRF sells cardboard and newsprint to buyers who recycle the materials
back into the original products, but the mixed paper stream, which cannot be recycled back into
one type of paper, is incorporated into a blend used to make cardboard.
Plastics are also hard to separate for closed-loop recycling. NIR units may be
programmed to identify individual types of plastics, however plastic packaging may contain
blends of plastics, or fillers and dyes that make the plastics harder for the NIR units to identify.
Mixed plastic fractions cannot be recycled back into packaging, but are used in lower-quality
applications such as barriers and park benches (Cimpan, Maul, et al. 2015). At the Spanish
MRF, mixed plastic and plastic bags were collected for incineration as refuse derived fuel

96
(RDF). The low grade of the PET and HDPE outputs in the modeled scenarios in Chapter 3
suggest that these output streams would be unsuitable for closed-loop recycling. On6 option to
improve the grade of the plastic outputs, as shown in Scenario 4, is to add quality control
personnel to the output steam to remove impurities, though this adds to the cost of sorting the
material. An MRF could also invest in new equipment, or tune the NIR units to be more
selective, at a cost to recovery. Finally, the MRF could sell the plastic to a secondary processor
where the plastic could be re-sorted and impurities removed. In the LCA, however, this step
would add an additional emissions burden.

4.1.3 LCA Models

We have shown that the calculations associated with life cycle assessments are sensitive
to assumptions, and require information about specific materials and how they are managed.
LCA is therefore usually carried out using modeling software in conjunction with a database of
material properties. There are several waste management LCA models used in the literature.
These include EASEWASTE 3 , developed by the Technical University of Denmark (Rigamonti
et al. 2014; Merrild, Larsen, and Christensen 2012), and the Ecoinvent database with SimaPro
or OpenLCA software (Huysman et al. 2015). An alternative model, the Solid Waste
Optimization Life Cycle Framework (SWOLF), is developed by Levis et al. (2013) for waste
management decision support under different scenarios. This work focuses on a fourth option,
the EPA's Waste Reduction Model (WARM). 4 The stated purpose of WARM is to help
municipalities and other decision makers quantify the climate impacts of materials management
decisions (U.S. EPA 2016e). Published applications of WARM include a comparison of the
GHG emissions from single and dual stream recycling systems, and an assessment of residential
curbside recycling programs in Florida (Fitzgerald, Krones, and Themelis 2012; Maimoun,
Reinhart, and Madani 2016).
WARM, a spreadsheet tool, is designed to compare emissions from different waste
management pathways to support voluntary GHG reporting by state and local solid waste
managers in the United States. The model uses GHG emission factors, developed using a life-

3 EASEWASTE has been replaced by EASETECH, the "Environmental Assessment System for Environmental
TECHnologies." See http://www.easetech.dk/EASEWASTE for more information about the model.
4 See https://www.epa.gov/warm for WARM documentation, and to download the Excel tool.

97
cycle assessment methodology, to calculate the GHG emissions and energy for a baseline and
alternative management scenario. The waste management practices in WARM include
landfilling, composting, combustion for energy recovery, recycling, and source reduction. The
results are calculated in terms of metric tons of carbon dioxide equivalent (MTCO 2 E) per short
ton of end-of-life material. Values reported here are in terms of metric tons of EOL material.

4.2 Material Recovery and GHG Emission Factors in WARM

The emission factors for recycling in WARM are based on two components. The first
is the difference in GHG emissions between manufacture of a product using virgin inputs or
recycled inputs. Calculating this number requires life cycle assessment, and is subject to the
uncertainties and assumptions described in the previous section. The second component of the
emission factor is the end-of-life recycling rate. This value adjusts the calculated savings
between virgin and recycled manufacture of a product from terms of recyclate into terms of
end-of-life material. This is necessary so the GHG emissions can be compared directly with
other management strategies for the same amount of waste. The EOL-RR is a function of the
recovery rate of materials in sorting facilities, so our experimental and modeled recovery data
can be used in this calculation.

4.2.1 Material Loss in Sorting and Manufacture

The EPA's values for retention rates of some materials in sorting and reprocessing, and
their product, the EOL-RR, are summarized in Table 4-3. The first column accounts for material
loss during collection and sorting. If we assume that collection losses are minimal, which is
reasonable for curbside collection systems, and that most mixed recyclables are sorted at an
MRF, which is consistent with EPA assumptions, then the first column corresponds with the
recovery of materials at an MRF.

98
Table 4-3: EPA WARM sorting, manufacturing, and overall retention rates for materials in MSW (U.S.
EPA 2016e)

Reprocessing End-of-life
Recovery rate retention rate recycling rate
(R) (P) (EOL-RR)
Recycled Material
Aluminum Cans 1.00 1 0.93 { 0.93
Steel cans 1.00 0.98 0.98
Glass 0.90 0.98 0.88
HDPE 0.92 0.93
i
0.86
PET 0.95 0.94 0.89
Corrugated Cardboard 1.00 0.93 0.93
Magazines/3rd Class Mail 0.95 0.71 0.67
Newspaper 0.95 0.94 0.89

A comparison of WARM recovery rates with MRF model results and experimental data
is presented in Table 4-4. The table also includes the equipment used to sort each material, and
its efficiency. The starred equipment and values are those for which we used literature sorting
efficiencies rather than experimental values. The fourth column of recovery values contains
modeled results for the light packaging recovery section (LPRS) of the Spanish MRF, which
has a different configuration from the scenarios modeled in Chapter 3, but relies on similar
equipment. These data are included because they are based on experimental data from a
functioning, state-of-the-art MRF, so are a good point of reference.

99
Table 4-4: Comparison of recovery rates for selected materials used in WARM with modeled and
measured recovery rates and target sorting efficiencies

Recovery _7

MRF MRF LPRS of Target


EPA Mode Optimization Spanish Sorting Equipment Sorting
WARM Scenario D MRF Efficiency

Aluminum 1.00 0.75 0.96 0.81 Eddy Current 0.80

Cardboard 1.00 0.73* 0.97 - Ballistic / Disc Screen* 0.76 / 0.75*

Ferrous 1.00 0.91 0.99 0.92 Magnet 0.92

Glass 0.90 0.93* 0.97* - Glass Breaker Screen* 0.97

HDPE 0.92 0.63 0.55 0.89 NIR 0.71

Paper 0.95 0.90* 0.94 - Ballistic / Disc Screen* 0.75 / 0.94*


PET 0.95 0.91 0.67 0.98 NIR 0.93

Plastic Film - 0.98* 0.98* - Vacuum* 0.98*

The data in Table 4-4 demonstrate that for most materials, there is a range of possible
recoveries depending on the sorting conditions. The target sorting efficiencies represent a
theoretical upper bound for target material recovery in a system without recirculation. For
example, the target material for the eddy current is aluminum. Our experiments showed that the
sorting efficiency of aluminum by the eddy current is 0.8. The maximum amount of aluminum
that can be recovered by series of sorting units that include one eddy current without
recirculation is therefore 80%. With a recirculation loop, the recovery rate may be higher than
the target sorting efficiency. This is observed in the LPRS of the Spanish MRF, which had two
recirculation loops, and in Optimization D, which had one.
Upstream losses due to incorrect non-target sorting may significantly degrade recovery
below that of a material's target sorting efficiency. Under realistic sorting conditions, the rate
of incorrect sorting of non-target materials is usually small but non-zero, and the order of sorting
equipment becomes important to maximize material recovery. In Optimization D, for example,
the NIR units sorting for PET and HDPE are the last two units before the recirculation loop.
The experimental sorting efficiencies for the NIR units are 0.93 for PET, and 0.71 for HDPE,
however the observed recovery of each material is much lower. This is due to incorrect sorting
of the plastics in the ballistic separator, which lies upstream from the NIR units. Our

100
experimental data from the Spanish facility showed that almost 20% of both PET and HDPE
entering the ballistic separator is incorrectly diverted into the 2D paper and cardboard stream.
Therefore, less than 80% of the plastic entering the sorting facility in Optimization D reaches
the target sorting units.
Although we have modeled sorting efficiencies as static for a given sorting technology,
the findings presented in Chapter 2 suggest that sorting efficiencies vary with the type and shape
of input composition. We do not have enough data to accurately model the relationship between
input waste composition and sorting efficiencies, however the tests to measure eddy current and
magnet sorting efficiency under varied input composition showed that while there was a
relatively consistent upper bound for recovery, under some conditions (e.g., flattened cans) the
recovery rate was severely reduced (see Chapter 2.4). These findings suggest that for some
input composition scenarios, material recovery in an MRF may be below the values in Table
4-4.
The modeled recovery for glass stands out as being higher than the WARM recovery
estimate, however our model results are based on a literature value for the sorting efficiency of
the glass breaker screen. This sorting technology breaks glass bottles and then sifts out the small
pieces of broken glass to separate them from other types of waste. Based on our conversations
with operators of both the Spanish MRF, and the MRF in Massachusetts, our understanding is
that most glass bottles are broken before they enter the MRF, and are therefore mixed in with
the fine debris separated by the scalping screen. This not only impacts the recovery rate, but
also significantly degrades the quality of the recovered glass, which makes the assumption of
closed loop recycling invalid.
The material recovery rates used in WARM to calculate GHG emission factors for
recycling may not reflect realistic losses in the sorting process. The objective of this comparison
is not to discount the recovery rates assumed in WARM, as they may be valid under some
conditions, but rather to introduce the MRF model as a tool for obtaining better recovery
estimates. The documentation for WARM states "There is uncertainty in the loss rates: some
material recovery facilities and manufacturing processes may recover or use recycled material
more or less efficiently than as estimated here" (U.S. EPA 2016e, 2-6). The MRF model
provides a means of estimating material recovery rates for specific conditions, and in
conjunction with an LCA tool like WARM, could help to improve the accuracy of LCA.

101
4.2.2 Application of Model Results to GHG Emission Factors

We found earlier that, according to EPA estimates, making aluminum cans out of one
ton of recycled ingots instead of virgin material saves 10 metric tons of CO 2 equivalent
emissions. Using the EPA's end-of-life recycling rate of 0.93, this equates to 9.3MTCO 2E saved
per ton of aluminum cans recycled. Using our modeled aluminum recovery of 0.75 from
Scenario 2 instead of the 1.0 from WARM, the modified end-of-life recycling rate is
EOLRRAI = (0.7 5)(0.93) = 0.70
and recycling one ton of aluminum cans yields a savings of 7.0 MTCO 2 E, 25% less than the
original estimate.
In 2014, Americans recycled 0.63 million tons of aluminum. Using the WARM EOL-
RR estimate, this resulted in a GHG benefit of 5.9 MMTCO 2E. Under the modified EOL-RR
of 0.7, the GHG benefit would be 4.4 MMTCO 2E. Applying the EPA's estimate for vehicle
emissions, the difference between these two values, 1.5MMTCO 2E, is equivalent to the annual
emissions of about 300,000 cars.
Virgin aluminum has a very high embodied energy, so the benefits from recycling
aluminum are, depending on assumptions, clear. For other materials, including plastics, the
benefits from recycling are smaller, and the impact of lower material recovery rates potentially
more decisive. For example, if an MRF has very poor recovery of plastics, as in Optimization
D, then using the reprocessing retention rates in Table 4-3, the EOL-RR for PET and HDPE
will be EOLRRPET = (0.67)(0.94) = 0.63 and EOLRRHDPE = (0.55)(0.93) = 0.51 . The GHG

benefits from recycling one ton of each plastic is then 0.87 MTCO 2E for PET, and 0.57
MTCO 2E for HDPE. The net emissions from combustion, in comparison, are estimated to be
1.33 MTCO 2E per ton of PET combusted, and 1.36 MTCO 2E per ton of HDPE combusted.
Recycling these plastics still provides an environmental benefit over both landfill and
incineration, but due to the lower material recovery rates, the benefits are smaller by 40% for
HDPE and 30% for PET than if calculated based on recovery rates in WARM. A waste manager
could use this information to either improve the sorting of plastics in her facility, or, particularly
if the amount of plastic in the waste stream were small relative to other recyclables, could
allocate resources to improving the recovery of a different material that would yield higher
GHG benefits.

102
4.2.3 Limitations to Using the MRF Model in Life Cycle Analysis

The availability and quality of waste data is a primary limitation both to the MRF model
in general, and to its application for LCA. To accurately represent a sorting facility, and
therefore to provide reliable estimates of recovery and grade, the model requires information
about the composition of the input waste stream, the sorting efficiencies of the equipment, and
the organization of the plant and personnel. Our work with the Spanish MRF demonstrated that
obtaining these data, particularly reliable measurements of waste composition, is a difficult and
expensive task. Given resource constraints, it is unlikely that municipalities in the United States
would be able to obtain the necessary data to tailor the model to their location.
An additional challenge for LCA is the granularity of material data. For our work with
the Spanish MRF, we established 14 material categories for waste stream characterization. An
initial list contained 24 categories, but we concluded that using the shorter list would meet our
needs for the model and greatly reduce the time required to hand-sort each waste sample into
its material components. A consequence of this decision is that our material categories are
broad, and may not provide adequate information for the LCA. As an example, we lumped
paper and cardboard into one material category, because they are sorted together in the Spanish
MRF. As the discussion of closed and open-loop recycling demonstrates, however, the
distinction between office paper, corrugated cardboard, newsprint, and other types of paper is
very important for determining downstream uses and environmental benefits.
Faster and lower-cost acquisition of waste stream data is therefore an important area
for future work. Mechanized or automated visual systems to measure the composition of waste
will need to be developed. Existing technology could provide some solutions. For example, the
detector in an NIR unit scans the materials passing through it on the conveyor and, depending
on the library of spectra in use, identifies most of the material types as well as the area of the
conveyor they cover. This information could be translated into waste composition data,
however challenges remain to correlate area with mass and three-dimensional shape.
There are other factors that influence how waste is managed that are not currently part
of either the MRF model or LCA. One such factor is the local materials market, which may
dictate how recovered materials are used even if LCA results point to a better alternative. For
example, the operators of the Massachusetts MRF report that the PET fraction from their plant
is made into carpeting and clothing, though the quality of the recovered plastic is high (above

103
97%). The colored HDPE, such as that used in laundry detergent bottles, is also sorted with a
high purity, but is made into products such as car dashboards, bumpers, and children's toys.
Local market forces make these the best buyers for this particular MRF, even if an
environmental analysis would argue in favor of closed-loop recycling whenever possible.

4.3 Implications for Recycling Policy

The United States does not have a national recycling policy. The Resource Conservation
and Recovery Act (RCRA), passed by congress in 1976, grants the EPA authority to establish
a framework for solid waste control in the United States. Non-hazardous solid waste is governed
by Subtitle D of RCRA, which includes provisions such as a ban on open dumping of waste,
and minimum criteria for municipal and industrial landfills. Implementation of RCRA
requirements for non-hazardous waste, including recycling, is left to states (U.S. EPA 2017).
As a result, recycling policies vary widely across the country, and as there are no national
reporting requirements, information about recycling programs across the country must be found
on a state-by-state basis, or collected by industry or interest groups. For example, the Northeast
Recycling Council reports that in 2017, twenty-two states had at least one mandatory recycling
requirement (though for some states this is limited to mandatory recycling of lead acid
batteries), and ten states had bottle bills (Northeast Recycling Council 2017)
The status of recycling policy in the United States contrasts sharply with the European
Union. EU member states are responsible for waste management, but the European
Commission's Waste Framework Directive codifies EU-wide principles for waste management
and recycling. These includes the Polluter Pays Principle and Extender Producer Responsibility
(European Comission 2008). The EU furthermore has packaging-specific legislation that sets
targets for recycling of certain materials in packaging (European Comission 2017).
Without a national recycling policy in the United States, improved tools for modeling
waste flows and estimated GHG benefits will be used on a state and local level. A national
carbon tax would generate greater scrutiny and development of LCA tools such as WARM,
which could be used for mandatory, rather than voluntary, GHG reporting, but at present a price
on carbon is unlikely. We consider three possible users of our model under current U.S. policy,

104
and identify areas where policy drivers are needed to advance sustainable materials
management.

4.3.1 MRF Model Users and Incentives

The MRF model enables estimation of recovered material quality and quantity, which
can inform decisions about how to balance the financial and environmental costs and benefits
of waste management. We have discussed sources of uncertainty for both the MRF model and
waste LCA, and decision makers who use these models need to be aware of the limitations. We
envision three primary users of the MRF model: MRF operators, local waste managers, and
local or state policy makers. Each of these users could gain useful information from the model,
but implementation of model results will require incentives that are currently lacking.
MRF operators seeking to improve profits or meet grade requirements can use the MRF
model to evaluate changes in design and operating parameters. While there are financial
incentives for MRFs to improve some aspects of operations, such as recovery of valuable
materials for sale, the model reveals some possible conflicts with environmental goals. For
example, most MRFs have to pay to send materials to landfill. Furthermore, as long as outputs
meet grade requirements, then profits are improved by increasing the total mass of material in
the outputs. This could create a disincentive for MRFs to sort residue efficiently, as
incorporating it into valuable outputs both increases the selling price of those outputs (assuming
that the minimum grade target is met), and reduces the mass of material sent to landfill. A
possible policy solution to this could be to set recovery targets for landfill as well as valuable
outputs. The overall sorting efficiency metric accounts for both recovery of valuable materials
and residue, so providing an incentive to improve that metric could help align incentives.
The second proposed user of the MRF model is local or regional waste managers.
Modeling the efficiency of individual MRFs could help divert recycling streams from different
sources to the facilities where they would be sorted best. Local managers could also, as the EPA
suggests, use WARM to calculate the environmental impacts of waste management. Here,
however, there is a distinct lack of policy incentive in the United States, as GHG reporting is
voluntary. Waste management is paid for on a local level, so unless GHG emissions are
regulated and priced, financial costs, rather than environmental concerns, will drive local
management decisions.

105
The third proposed group of MRF model users are policy makers themselves at local
and state levels. Without knowledge of how materials are sorted, recycling policies may not
achieve desired outcomes. For example, Massachusetts has a state-wide ban on landfilling or
combusting certain recyclable materials including glass and metal containers, single polymer
plastics, and recyclable cardboard and paper (Massachusetts Department of Energy and
Environmental Affairs 2014). Our work with MRFs showed that glass is very difficult to
separate from comingled streams, and may degrade the quality of other outputs. Furthermore,
that almost half of the glass from single stream recycling is used in open loop recycling, where
the benefits from avoided material use are low. Collection of glass in single stream recycling
may therefore have an environmental cost rather than a benefit.

106
Chapter 5
Conclusion
In this work, we applied a network flow model of material recovery facilities to compare
the performance of different realistic sorting facilities and to find configurations that improve
sorting efficiency and profit. We conducted experiments to measure the variation of sorting
equipment efficiency under different operating conditions. Finally, we discussed the role that
MRFs play in recycling systems, and showed how material recovery rates are applied to life
cycle assessment.

5.1 Sorting Efficiencies and Variation

Previous MRF models have focused on facility energy use and profit, but lack of
available waste stream composition and sorting efficiency data limited the ability to model
material flows. This work presents experimental sorting efficiencies obtained at a functioning
MRF. These efficiencies enable a greater level of detail in material flow modeling, as the fate
of all materials at each sorting step may be determined. Chapter 2 presents experimental data
for single stream composition and MRF sorting efficiencies, as well as measured sorting
efficiency sensitivity to input composition and equipment parameters. These data demonstrate
that non-target sorting efficiencies are usually nonzero, and that defining them is critical to
calculating grade. We also show that real sorting efficiencies are variable and depend on both
input composition and equipment settings.

5.2 Optimization of MRF Design

The optimization function of the model finds the best performing configurations to
maximize a weighted combination of profit and sorting efficiency. We used the model to find

107
improvements to an MRF, and showed that optimizing the configuration improves sorting
efficiency by 9%. The addition of a recirculation loop yielded an additional 8.6% improvement
in sorting efficiency.

5.2.1 Pricing and Design for Profit

We modeled an MRF with and without quality control, and found that adding quality
control to three outputs, and thereby improving the grade of those outputs, slightly decreased
the calculated profits. The QC reduced the mass of the recovered outputs, but did not increase
their value since they already met the quality requirement. Pricing schemes that use a fixed
price above a grade threshold may therefore create incentives for MRFs to maximize the
allowable contamination in the output. If maximizing grade is desirable, then a pricing scheme
that uses a linear relationship between grade and price without a threshold is necessary. This
could improve the overall efficiency of the recycling system by reducing the need for
downstream processing.

5.3 Recovery in LCA

Life cycle assessment is a useful tool for estimating the benefits of choosing one waste
management strategy over others. The results of LCA are based on a series of assumptions, one
of which is the recovery rate of materials in the sorting process. The MRF model provides a
means of finding realistic recovery estimates, which can be applied to LCA. We show that
modeled recovery rates are, for most materials, lower than the estimates used in the EPA's
waste LCA model. As recovery rates are used to normalize LCA findings to units of end-of-life
products, they have a direct impact on the magnitude of calculated benefits.

5.3.1 Implications for Policy

The MRF model may be used by plant managers to improve profits, but larger scale
implementation of model insights, particularly changes to recycling systems to improve GHG
savings, will require policy incentives that are currently lacking in the United States.

108
Municipalities bear the costs of waste management, including recycling, and without regulatory
pressure, only changes that reduce local costs will be implemented.

5.4 Future Work

This thesis builds on previous work that developed and validated the MRF model.
Although the data presented here demonstrates the utility of the model in evaluating MRF
scenarios to inform design decisions, the application to real MRFs is limited by the availability
of data on input stream composition and sorting efficiency variability. In addition, a more
complete model of pricing for materials and purity requirements would improve the profit
calculation.

5.4.1 Automated Waste Composition Measurement

Given the challenge associated with collecting waste composition data by hand, future
efforts to improve MRF modeling should focus on developing automated systems to
characterize the materials in a waste stream. Proposed methods for gathering waste data include
vision systems coupled with computer programs that can be trained to recognize materials.
Faster and cheaper collection of data would allow the model to be quickly tailored to a specific
MRF, where it could be used to inform facility-scale decisions. Two areas where automated
data collection would be particularly helpful are measuring waste composition variations over
time, and expanding the number of material categories. The latter effort has implications for
recycling LCA, as specific material categories follow different paths for recycling.
The experiments presented in this thesis demonstrate that there are relationships
between input composition, sorting equipment parameters, and sorting efficiencies. The current
model assumption that sorting efficiencies are static is therefore a source of error, and defining
how sorting efficiencies depend on other factors an important means of improving the model.
Given the complexities of heterogeneous waste, this effort would also rely on automated
systems to measure the composition of waste streams in and out of sorting equipment.

109
5.4.2 Model of Price Data for Varied Grades of Material

Our model assumes a linear relationship for selling price and grade between a
concentration threshold and the purity requirement of an output. This relationship is based on
the market for metals, and may not apply well to other materials, or to MRFs in areas with
limited markets. A better understanding of how the grades of MRF outputs relate to selling
prices would improve the profit calculation.

110
Appendix

A.1 Mixed Waste MRF in Spain

main feeder
cardboard

. ............. foLmb.e
la n d i ll ....

-fdfl bulky waste ----- .1ib n . .

...
-.-.-........
.nd....

ferrous magnet me without-

biologic treatment i ae ador

bag opnr

scra

feus

n
Equpmntpscese

PET HDPEpaper & cardboard

- - - - - -- -- - ferrous

aluminumrefuse derived fuel


landfill

Equipm;en processes
Mechanical Sorting Unit material without changing
the composition

. ahual 'Srt1ngSfafien. Output Collection Unit


E . - -7

111
A.2 Output Requirements for Spanish MRF
The sum of the materials separated by semi-colons cannot exceed the given maximum
concentration. Data is based on Ecoembes rec uirements, as provided to us by the Spanish MRF
Output Materials in Output Concentration Notes
Min ax
Should also be non-
0.80 1.00
Aluminum ferrous metals <0.5%
Ferrous 0.00 0.00 Specified as zero
HDPE; PET; PP; Film;
OtherPlastics; Tetrabrik; 0.00 0.04
PaperAndCardboard
Aluminum
HDPE; PET; PP; Film; 0.00 0.02
OtherPlastics
PaperAndCardboard 0.00 0.02
Tetrabrik 0.00 0.02
Fines; Other; Organic; Glass;
0.00 0.06
Inerts0
Ferrous Ferrous 10,80 1.00
HDPE 0.85 1.00
PET; PP; OtherPlastics; Film 0.00 0.10
Based on Ecoembes
HDPE Ferrous; Aluminum 0.00 0.01 requirements for color
PaperAndCardboard; HDPE
Tetrabrik; Fines; Organic; 0.00 0.05
Glass; Inerts; Other
PP and
0.00 1.00
OtherPlastics PP; OtherPlastics
PET 0.92 1.00 Color and Natural
Ferrous; Aluminum 0.00 0.01
PET HDPE; PP; OtherPlastics;
Film; Fines; Organic;
0.00 0.07
PaperAndCardboard; Also 0.5% maximum
Tetrabrik; Glass; Inerts; Other for PVC
Tetrabrik 0.95 1.00
HDPE; PET; Ferrous;
0.00 0.03
Aluminum "Other containers"
Tetrabrik Fines; Organic;
PaperAndCardboard;
0.00 0.02
OtherPlastics; PP; Film;
Glass; Inerts; Other
Paper and 0.97 1.00
Cardboard
_PaperAndCardboard

112
A.3 Best Design for Light Packaging Recovery Section to Maximize
Profit

Best LPRS Config uration Input from


Profit: 824 Euros,/hr Plant
Efficiency: 94.6%
> MagnetFeru

Eddy Current

HDPE NIR 3

NIR 1

Tetra Pa k
NIR 4 NIR 6

NIR 2 Landfill

PET Output

S Magnet Otu

NIR 1

Original LPRS Configuration


PET NIR 2 NIR3
Profit: 766 Euros/hr
Efficiency: 95.0%
Teta akNI4 NIR5 lstc

Redmulation :Plastics M, - TTetrPak NIR 6 Eddy Crrent

113
114
References
Ashby, Michael. 2013. Materials and the Environment: Eco-Informed MaterialChoice. 2nd ed.
Waltham, MA: Elselvier Inc.

Bjbrklund, Anna, and Gbran Finnveden. 2005. "Recycling Revisited - Life Cycle Comparisons of
Global Warming Impact and Total Energy Use of Waste Management Strategies." Resources,
Conservation and Recycling 44 (4): 309-17. doi:10.101 6/j.resconrec.2004.12.002.

BLS Magnet. 2017. "Magnetic Separation, Image of Magent." Accessed May 20.
http://www.blsmagnet.com/en/products/industrial-applications/magnetic-separation.

Chester, Mikhail, Elliot Martin, and Nakul Sathaye. 2008. "Energy, Greenhouse Gas, and Cost
Reductions for Municipal Recycling Systems." Environmental Science & Technology 42 (6).
American Chemical Society: 2142-49. doi:10.1021/es0713330.

Cimpan, Ciprian, Anja Maul, Michael Jansen, Thomas Pretz, and Henrik Wenzel. 2015. "Central
Sorting and Recovery of MSW Recyclable Materials: A Review of Technological State-of-the-
Art, Cases, Practice and Implications for Materials Recycling." JournalofEnvironmental
Management 156: 181-99. doi:10.1016/j.jenvman.2015.03.025.

Cimpan, Ciprian, Marianne Rothmann, Lorie Hamelin, and Henrik Wenzel. 2015. "Towards Increased
Recycling of Household Waste: Documenting Cascading Effects and Material Efficiency of
Commingled Recyclables and Biowaste Collection." JournalofEnvironmental Management 157
(July): 69-83. doi:10.1016/j.jenvman.2015.04.008.

Combs, Anthony Rush. 2012. "Life Cycle Analysis of Recycling Facilities in a Carbon Constrained
World." North Carolina State University.

Edwards, D.W., and J. Schelling. 1996. "Municipal Waste Life Cycle Assessment Part 1, and
Aluminum Case Study." Process Safety and Environmental Protection74 (Part B): 205-22.

European Comission. 2017. "Packaging and Packaging Waste." Accessed February 26.
http://ec.europa.eu/environment/waste/packaging/legis.htm.

. 2008. "Directive 2008/98/EC on Waste." Official Journalof the European Union L 312: 3-
30. http://eur-lex.europa.eu/legal-
content/EN/TXT/PDF/?uri=CELEX:32008L0098&qid=1483993056768&from=EN.

Fitzgerald, Garrett C., Jonathan S. Krones, and Nickolas J. Themelis. 2012. "Greenhouse Gas Impact
of Dual Stream and Single Stream Collection and Separation of Recyclables." Resources,
Conservation and Recycling 69 (December): 50-56. doi:10.101 6/j.resconrec.2012.08.006.

Graedel, Thomas E. 2011. "Recycling Rates of Metals: A Status Report."


http://www.unep.org/resourcepanel/Portals/24102/PDFs/Metals_RecyclingRates _110412-1.pdf.

Huysman, Sofie, Sam Debaveye, Thomas Schaubroeck, Steven De Meester, Fulvio Ardente, Fabrice
Mathieux, and Jo Dewulf. 2015. "The Recyclability Benefit Rate of Closed-Loop and Open-
Loop Systems: A Case Study on Plastic Recycling in Flanders." Resources, Conservationand
Recycling 101. Elsevier B.V.: 53-60. doi:10. 1016/j.resconrec.2015.05.014.

115
International Energy Agency. 2016. "World Energy Outlook 2016." www.iea.org.

Ip, Karine, Mariapaola Testa, Anne Raymond, Stephen Graves, and Timothy Gutowski. 2016.
"Performance Evaluation of Material Separation in a Material Recovery Facility Using a
Network Flow Model." Submittedfor Publication.

Jaunich, Megan K., James W. Levis, Joseph F. DeCarolis, Eliana V. Gaston, Morton A. Barlaz,
Shannon L. Bartelt-Hunt, Elizabeth G. Jones, Lauren Hauser, and Rohit Jaikumar. 2016.
"Characterization of Municipal Solid Waste Collection Operations." Resources, Conservation
and Recycling 114. Elsevier B.V.: 92-102. doi: 10.10 16/j.resconrec.2016.07.012.

Koermer, Scott Carl. 2015. "The Application of Mineral Processing Techniques to the Scrap
Recycling Industry." Virginia Polytechnic Institute and State University.

Levis, James W., Morton A. Barlaz, Joseph F. DeCarolis, and S. Ranji Ranjithan. 2013. "A
Generalized Multistage Optimization Modeling Framework for Life Cycle Assessment-Based
Integrated Solid Waste Management." Environmental Modelling & Software 50: 51-65.
doi:10.1016/j.envsoft.2013.08.007.

Maimoun, Mousa A., Debra R. Reinhart, and Kaveh Madani. 2016. "An Environmental-Economic
Assessment of Residential Curbside Collection Programs in Central Florida." Waste
Management 54. Elsevier Ltd: 27-38. doi:10.1016/j.wasman.2016.04.025.

Massachusetts Department of Energy and Environmental Affairs. 2014. Solid Waste Facility
Regulations. http://www.mass.gov/eea/agencies/massdep/recycle/regulations/31 0-cmr- 19-
00.html.

Merrild, Hanna, Anna W Larsen, and Thomas H Christensen. 2012. "Assessing Recycling versus
Incineration of Key Materials in Municipal Waste: The Importance of Efficient Energy Recovery
and Transport Distances." Waste Management 32: 1009-18. doi:10.1016/j.wasman.2011.12.025.

Miranda, Ruben, M. Concepcion Monte, and Angeles Blanco. 2013. "Analysis of the Quality of the
Recovered Paper from Commingled Collection Systems." Resources, Conservation and
Recycling 72: 60-66. doi:10.1016/j.resconrec.2012.12.007.

Northeast Recycling Council. 2017. "Disposal Bans & Mandatory Recycling in the United States."
Brattleboro, VT.
https://nerc.org/documents/disposalbansmandatoryrecycling-united-states.pdf.

Oregon Department of Environmental Quality. 2011. "Composition of Commingled Recyclables


Before and After Processing Prepared for the Oregon Commingled Recycling System
Improvement Workgroup." Portland, OR.
http://www.deq.state.or.us/lq/pubs/docs/sw/CompositionCommingledRecyclablesBeforeAfterPro
cessing.pdf.

Pressley, Phillip N., James W. Levis, Anders Damgaard, Morton A. Barlaz, and Joseph F. DeCarolis.
2015. "Analysis of Material Recovery Facilities for Use in Life-Cycle Assessment." Waste
Management 35. Elsevier Ltd: 307-17. doi: 10. 1016/j.wasman.2014.09.012.

116
R.W. Beck Inc. 2005. "Pennsylvania Recovered Material Composition Study."
http://files.dep.state.pa.us/waste/Recycling/lib/landrecwaste/recycling/documents/recmatcomp.
pdf.

Rigamonti, L., M. Grosso, J. Moller, V. Martinez Sanchez, S. Magnani, and T.H. Christensen. 2014.
"Environmental Evaluation of Plastic Waste Management Scenarios." Resources, Conservation
and Recycling 85. Elsevier B.V.: 42-53. doi: 10.1016/j.resconrec.2013.12.012.

Testa, Mariapaola. 2015. "Modeling and Design of Material Recovery Facilities: Genetic Algorithm
Approach." MIT.

Tinsley Company. 2017. "Ballistic Separators." Accessed May 20.


http://www.tinsleycompany.com/recycling/ballistic-separators/.

Tomra. 2017. "Recycling Sorting Solutions: Autosort." Accessed May 20.


https://www.tomra.com/en/solutions-and-products/sorting-solutions/recycling/products/autosort/.

Turner, David A., Ian D. Williams, and Simon Kemp. 2015. "Greenhouse Gas Emission Factors for
Recycling of Source-Segregated Waste Materials." Resources, Conservation and Recycling 105:
186-97. doi:10.1016/j.resconrec.2015.10.026.

U.S. EPA. 2017. "Resource Conservation and Recovery Act (RCRA) Laws and Regulations."
Accessed March 4. https://www.epa.gov/rcra.

. 2015. "Advancing Sustainable Materials Management: 2013 Fact Sheet."


https://www.epa.gov/sites/production/files/2015-09/documents/2013_advncngsmmfs.pdf.

. 2016a. "Advancing Sustainable Materials Management: 2014 Fact Sheet."


https://www.epa.gov/sites/production/files/2016-11 /documents/2014_smmfactsheet_508.pdf.

. 2016b. "Advancing Sustainable Materials Management: 2014 Tables and Figures."


https://www.epa.gov/sites/production/files/2016-
1 1/documents/2014_smmtablesfigures_508.pdf.

. 2016c. "Documentation for Greenhouse Gas Emission and Energy Factors Used in the Waste
Reduction Model - Background Chapters." https://www.epa.gov/sites/production/files/2016-
03/documents/warm_vI 4background.pdf.

. 2016d. "Documentation for Greenhouse Gas Emission and Energy Factors Used in the Waste
Reduction Model - Containers, Packaging, and Non-Durable Goods Materials Chapters."
https://www.epa.gov/sites/production/files/2016-
03/documents/warm_vi 4_containers-packagingnon-durablegoodsmaterials.pdf.

. 2016e. "Documentation for Greenhouse Gas Emission and Energy Factors Used in the Waste
Reduction Model - Management Practices Chapters."
https://www.epa.gov/sites/production/files/2016-
03/documents/warm_vI 4management-practices.pdf.

. 2017a. "Inventory of U.S. Greenhouse Gas Emissions and Sinks: 1990-2015."


https://www.epa.gov/sites/production/files/2017-02/documents/2017_completereport.pdf.

117
. 2017b. "Sustainable Materials Management: Non-Hazardous Materials and Waste
Management Hierarchy." https://www.epa.gov/smm/sustainable-materials-management-non-
hazardous-materials-and-waste-management-hierarchy.

Vazzoler. 2017. "Metal Separation, Eddy Current Separator Image." Accessed March 30.
http://www.vazzoler.it/separation.php.

Wolf, M.I. 2011. "Modeling and Design of Material Separation Systems with Applications to
Recycling." MIT.

Zink, Trevor, Roland Geyer, and Richard Startz. 2015. "A Market-Based Framework for Quantifying
Displaced Production from Recycling or Reuse." JournalofIndustrial Ecology, 719-29.
doi:I0.11 11/jiec.12317.

118

Potrebbero piacerti anche