International Journal of Sustainable Engineering

ISSN: (Print) (Online) Journal homepage: www.tandfonline.com/journals/tsue20

AI-based decision support system for enhancing

end-of-life value recovery from e-wastes

Ehsan Simaei & Shahin Rahimifard

To cite this article: Ehsan Simaei & Shahin Rahimifard (2024) AI-based decision support system
for enhancing end-of-life value recovery from e-wastes, International Journal of Sustainable
Engineering, 17:1, 80-96, DOI: 10.1080/19397038.2024.2306293

To link to this article: https://doi.org/10.1080/19397038.2024.2306293

© 2024 The Author(s). Published by Informa

UK Limited, trading as Taylor & Francis
Group.

Published online: 29 Jan 2024.

Submit your article to this journal

Article views: 2463

View related articles

View Crossmark data

Citing articles: 1 View citing articles

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=tsue20
INTERNATIONAL JOURNAL OF SUSTAINABLE ENGINEERING
2024, VOL. 17, NO. 1, 80–96
https://doi.org/10.1080/19397038.2024.2306293

AI-based decision support system for enhancing end-of-life value recovery from
e-wastes
Ehsan Simaei and Shahin Rahimifard
Centre for SMART, Wolfson School of Mechanical, Electrical and Manufacturing Engineering Loughborough University, Loughborough, UK

ABSTRACT ARTICLE HISTORY

The growing diversity and volume of e-waste products are emerging as significant and complex Received 10 January 2023
challenges from environmental, economic, and recycling perspectives. This surge not only hampers the Accepted 12 December 2023
effective management and conservation of precious resources and strategically important materials KEYWORDS
(SIMs) but also intensifies the problem. Addressing this pressing global issue necessitates a holistic and AI visual recognition;
intelligent decision-making model capable of effortlessly handling the vast variety and volume of e-waste optimized fuzzy logic;
products. This paper introduces a novel multi-criteria decision support system (DSS) that combines AI decision support system;
visual recognition with a multi-fuzzy model to precisely determine the optimal End-of-Life (EoL) recovery advanced material
route. Additionally, the functionality of the proposed DSS is showcased through a pilot implementation separation; automated
and case study examples, emphasising the advantages of a fully automated, reliable, and efficient EoL material recovery; e-waste
management for e-waste. recycling solution

Introduction
separation and recycling (Parajuly et al. 2016). While the
The proliferation of manufactured digital and non-digital arte­ ramifications of the fourth industrial revolution and other
facts in the 21st century has led to significant over-extraction digital transformation initiatives on resource management
and depletion of natural resources. Simultaneously, the disman­ remain ambiguous, embedding sophisticated electronic sys­
tling and decomposition of these products have become increas­ tems in nearly every domestic and industrial product could
ingly challenging due to their varied shapes, sizes, and inflate manufacturing costs and exacerbate the depletion of
complexities. To ensure the preservation of our planet for future essential resources (Schwab 2015). The pressing issue now is
generations, while also supporting economic growth and inno­ how to efficiently recycle the multitude of electronic products
vative production, a circular approach to material usage was that have reached, or will soon reach, their End of Life (EoL).
introduced in the 1970s as an alternative to traditional linear Current recycling methods lack the capability and quality
concepts and has evolved since then (Andersen 2007). needed to consistently retain precious metals (Herrmann
Several countries and regional treaties, including the et al. 2012).
European Union Commission (European Commission The idea of enhancing the concentration of Strategically
Environment 2011), have advocated for the circular economy Important Materials (SIMs) before large-scale fragmentation
model as a critical method to preserve all materials, including of e-waste has been presented and refined with the develop­
rare and precious elements. Nevertheless, rules and regulations ment of various tools and technologies (Jie, Barwood, and
alone cannot address the challenges of e-waste recycling. Rahimifard 2015). However, the stark reality of today’s
Enabling technologies, such as robust decision-making tools e-waste management is that neither manual (traditional) nor
and reliable automation, are essential (Saidani et al. 2020). robotic (modern) methods are individually sufficient to handle
Derived from data in a report by the United Nations (Forti the vast volume and diverse nature of e-wastes. Using a real-
et al. 2020), Figure 1 reveals that over 53.6 million tonnes of world scenario for context, consider the manufacturing of
e-waste were generated globally in 2019. Of this staggering Electric Vehicles (EV). One of the primary constraints is
amount, Europe alone produced 12 million tonnes, equating resource depletion, which is why EV production currently
to an average of 16.2 kg per person. Intriguingly, while Europe represents only a small percentage of total vehicle output. By
is home to less than 10% of the world’s population, it accounts ensuring efficient and automated SIM concentration and
for 28% of the total e-waste. What’s even more alarming is that recovery across all e-waste types, we can address this break in
current data indicate that less than 35% of e-waste is collected resource management and avert material exhaustion. This, in
and recycled in Europe (Eurostat 2021) turn, can pave the way for more sustainable EV manufacturing
The primary barriers to achieving a closed-loop material (Li, Barwood, and Rahimifard 2018).
flow are the absence of effective collection programmes and In light of these challenges, this paper introduces an AI-
the requisite advanced technological infrastructure for precise driven decision support system, aiming to showcase the

CONTACT Shahin Rahimifard s.rahimifard@lboro.ac.uk Centre for SMART, Wolfson School of Mechanical, Electrical and Manufacturing Engineering
Loughborough University, Loughborough LE113TU, UK
© 2024 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis Group.
This is an Open Access article distributed under the terms of the Creative Commons Attribution-NonCommercial License (http://creativecommons.org/licenses/by-nc/4.0/), which permits
unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited. The terms on which this article has been published allow the
posting of the Accepted Manuscript in a repository by the author(s) or with their consent.
 INTERNATIONAL JOURNAL OF SUSTAINABLE ENGINEERING 81

Figure 1. Total amount of e-wastes generated (left) and e-wastes generated per person (right) in 2019 within regions and global scale, derived from data published by
UN report (Forti et al. 2020).

potential of improved recovery and recycling strategies for commonly linked to Li et al. (2018). They described the cur­
future urban mining from e-waste products. We commence rent vehicle recycling processes as basic, largely ineffective in
with a review of pertinent literature and an exploration of the identifying and extracting valuable materials due to informa­
AI-based decision support system’s foundational concepts. tional voids and physical limitations. Their proposed metho­
Subsequent sections detail a pilot implementation of the pro­ dology leverages detailed product design information to guide
posed system, followed by a case study. Together, these sec­ disassembly planning, thereby enhancing prospects for more
tions converge on key insights and highlight avenues for future automated operations.
research. Other studies have taken a different angle, presenting math­
ematical solutions for disassembly sequencing and planning
(Bentaha, Battaiä, and Dolgui 2015; Tumkor and Senol 2007;
Background Zhu and Roy 2015). Behdad et al (2012, 2014). delved into
mathematical modelling for disassembly sequencing in the
Recent studies have underscored the potential of innovative early design stages. They used a stochastic programming
technologies and tools for End-of-Life (EoL) management of model to determine sequence and disassembly time, taking
e-wastes. These encompass a range, from automation solu­ into account potential damage to parts as a primary uncer­
tions, pre-planning, and redesigning products for improved tainty factor. In a shift from these approaches, Yeh and Wei
environmental impact, to the use of robotics and AI for pro­ (2013) proposed parallel disassembly over traditional sequen­
duct disassembly and disintegration. tial methods for managing End-of-Life waste products. Several
In the realm of recycling automation, the proposed solu­ researchers have explored disassembly sequencing, design for
tions often depend upon the specific products analysed in case reuse, and design for disassembly, often assuming that disas­
studies. For instance, Kim et al. (2009) put forward a fully sembly simply mirrors the assembly process (de Almeida,
automated robotic cell featuring mobile robots and a multi- Borsato, and Lie Ugaya 2017; Jin and Tian 2012; Radaschin
agent system. Meanwhile, the interplay between human-robot et al. 2012; Yi et al. 2008; Zülch and Hrdina 2010; Zülch and
collaboration in disassembly tasks, covering areas such as Schwarz 2008). The non-destructive removal of screws, nuts,
ensuring a safe environment, establishing communication pro­ and bolts is another approach that’s been explored by some
tocols, and sharing algorithms, has been the subject of explora­ authors, though their focus has mainly been on specific models
tion by several researchers (Bogue 2019; Kopacek 2005; or case study products (Friedrich, Lechler, and Verl 2016;
Vongbunyong, Vongseela, and Sreerattana-Aporn 2017). In Vongbunyong, Pagnucco, and Kara 2016).
a semi-automated collaborative approach proposed by Ullah The fuzzy decision-making technique has largely been
and Sarkar (2020), RFID and barcode tags are employed to used to study the management of e-waste reverse logistics
bolster the tracking and control of the disassembly process, (before disassembly) or at the material recycling stage
thus minimising the chances of e-wastes (or other wastes) (after disassembly). This approach aims to lessen envir­
being lost amongst other disposables. However, the practicality onmental impacts and bolster economic viability (Grimes
and reliability of such approach, especially in semi or fully and Maguire 2020; Kinoshita et al. 2020; Shumon,
destructive de-manufacturing contexts, remain underex­ Ahmed, and Ahmed 2016). Zhang et al. (2014) suggested
plored. The question of whether these tracking systems can a combination of parallel disassembly sequencing and
endure throughout a product’s life also hangs in the balance. In fuzzy rough sets for non-destructively disassembling com­
a distinct study by Liu et al. (2019), the dynamic between plex products. Likewise, fuzzy clustering has been
humans and robots within a disassembly cell was probed, employed to create cost models for disassembly (Ziqiang
utilising an array of Kinect cameras and computers to syn­ et al. 2015). In this approach, a mathematical model
chronously analyse and predict human and robot movements employs a fuzzy set to frame the issue, subsequently
in real-time. Turning attention to robotic disassembly for SIM evaluating the implications of disassembly costs through
concentration in Electric Vehicles (EVs), pivotal work is case studies.
82 E. SIMAEI AND S. RAHIMIFARD

Regarding the application of AI in End-of-Life manage­ Problem definition

ment, several academic publications have identified AI and
The review of related publications underscores the pivotal role
visual recognition as pivotal tools in e-waste recycling
that enhanced automation, decision-making, and visual recog­
operations (Mao et al. 2021; Vongbunyong, Kara, and
Pagnucco 2013). This includes using a single contour in nition systems play in optimising material recovery from
the sorting processes of e-waste recycling (Laszlo et al. e-waste. This research introduces an AI-based decision sup­
2019), where the recommended system employs a visual port system that uses visual recognition to strengthen material
recognition system to separate PC motherboard compo­ concentration before e-waste undergoes the fragmentation
nents that have been manually taken apart. In another process, aiming to boost the quality of recovered materials.
piece of research, the soar cognitive architecture was uti­ Present-day e-waste processing falls short, mainly due to
lised to identify and extract screws from various laptops a lack of a coherent strategy and a robust decision-making
(Difilippo and Jouaneh 2019). system. These systems are needed to identify the most suitable
Despite the multitude of publications discussing recovery options when products reach their EoL. As depicted
robotic and automated processing of wastes, the industry in Figure 2, current procedures for managing the end-of-life of
has been slow to adopt these innovations. One major e-wastes are largely non-automated, relying heavily on the
limitation stem from a general, yet flawed assumption judgement of low-skilled workers. The rapid proliferation of
within much of the research: that End-of-Life (EoL) pro­ digital products renders this manual approach ineffective. The
ducts can be approached as if they were brand new. vast amounts of e-waste accumulating in treatment centres,
However, the reality is that disassembly of used products paired with the digital industry’s challenges in accessing essen­
is far from being a mere reverse of the assembly process tial raw materials, spotlight an ineffective material recycling
during manufacturing. Throughout their use phase, and loop. This issue is further intensified by an over-reliance on
even during collection and handling, most EoL products shredding as a one-size-fits-all solution for all recycling pro­
undergo a range of physical damages. They exhibit blems (shown in Figure 3). Moreover, today’s e-waste assess­
changes in their original characteristics, endure unpredict­ ment methods, predominantly manual and performed on the
able wear and tear, and generally present conditions vastly spot by less-skilled workers, are slow and fall short of 21st-
different from those of new products. These variations century standards.
complicate their processing and disassembly, demanding The proposed AI-based decision support system, detailed in
a more nuanced and adaptive approach than what many Figure 4, promises a structured methodology. It will utilise
current methodologies offer. The true challenge lies in state-of-the-art automation to improve SIM concentration
developing an innovative solution tailored to the unique before fragmentation. This strategy ensures e-waste items are
challenges of used products. sorted by their material content and recovery value before

Figure 2. Current material flow from raw to parts and products back again to raw/energy.

Figure 3. Material flow within recycling phase (magnified from Figure 2).
 INTERNATIONAL JOURNAL OF SUSTAINABLE ENGINEERING 83

Figure 4. Brief overview of proposed material flow in connection with other recycling entities (to replace the process shown in Figure 3).

processing. This study operates under the assumption that identification step, the broader VEIDD framework takes over.
e-wastes have already been transported to treatment facilities, It delves into a detailed evaluation of the product, scrutinises its
without prior sorting or labelling. It’s also based on the pre­ current state of intactness, and estimates the complexity of
mise that a variety of EoL treatments are available at these extracting valuable components. This estimation considers ele­
facilities. Such an assumption provides the decision system ments like product design and how closely its parts are
with the flexibility needed to handle different e-waste types. assembled.
It’s crucial to note that this system primarily aims to recover Figure 5 provides a visual representation of VEIDD’s
high-value e-waste. Remaining and lesser value products can functionalities and how its sub-systems interplay. What
still be processed via traditional recycling methods, including sets VEIDD apart is its inherent flexibility and its adapt­
shredding. ability across various recycling processes. It offers a fresh,
more efficient approach compared to traditional, labour-
AI-Based decision support system to enhance EoL heavy recycling methods. VEIDD encompasses four main
management of e-wastes interconnected sub-systems, all functioning within an over­
arching automatic platform. Once the product is recog­
Disassembly processes, unlike their assembly and manufactur­ nised by the AI system, a user-established e-waste
ing counterparts, confront the uncertainties and unpredictabil­ database, which is continuously updated via online bots,
ity of End-of-Life products. This encompasses challenges like calculates the potential value embedded within the e-waste.
varying product conditions, the unpredictable availability of This valuation is a crucial step, paving the way for deter­
valuable parts, and discerning the most efficient extraction mining the most economical recovery approach. The sys­
methods. Decision-making in these uncertain environments tem further gauges the intactness of the product by
requires a detailed evaluation of numerous scenarios, where weighing it, which helps detect missing components, and
each action could lead to a spectrum of outcomes, thereby if the batteries are still connected. The next step assesses
assessing the risks (Nascimento 2019; Polat, Capraz, and potential processing challenges, factoring in the product’s
Gungor 2018). In contrast, during manufacturing, every com­
size and the intricacy of its design. Gathering insights from
ponent and subassembly is systematically labelled, tagged, and
each of these evaluations, VEIDD then consolidates this
maintained in peak condition. They are precisely timed and
data to inform the final decision on the optimal e-waste
positioned into the production line, minimising uncertainty,
management strategy.
and the risk of errors. This meticulous process means that
Following this overview, the subsequent subsections delve
surprises or unplanned events are rare in manufacturing.
deeper into the intricacies of VEIDD’s individual sub-systems.
However, recycling, especially e-waste recycling, doesn’t have
Each segment offers detailed insights into their operational
this luxury. Often, it’s nearly impossible to predict the type,
mechanics, functionalities, and roles within the broader frame­
condition, or the timing of items being sent for recycling, mak­
work providing comprehensive understanding of how these
ing the task of pre-planning for material recovery complex.
sub-systems synergise, ensuring VEIDD’s capabilities in
To address this challenge, we’ve crafted a flexible and auto­
e-waste management.
mated decision support system. Named VEIDD (standing for
Visual recognition of parts, Evaluation of estimated recovery
value, Intactness analysis, Difficulty assessment in processing,
Multi-criteria and fuzzy system selection
and support for final recovery Decisions), this system har­
nesses AI to streamline and enhance the material recovery The VEIDD system stands out largely due to its distinct
process. VEIDD aims to bolster the efficient retrieval of approach to multi-criteria selection, designed to seamlessly
Strategically Important Materials (SIMs) from e-wastes, tap­ integrate with a broad spectrum of e-waste. An underlying
ping into the prowess of advanced techniques, like robotic principle of this system’s design is its independence from any
disassembly, all in alignment with circular economy principles. specific manufacturer’s data. This design principle is crucial, as
The standout component of VEIDD is its AI-driven visual in the real-world landscape of e-waste management, discarded
recognition system. This tool is specifically designed to pinpoint products might have been on the market for an extended
the exact type and model of e-waste products. After this period, or they may originate from manufacturers who have
84 E. SIMAEI AND S. RAHIMIFARD

Figure 5. VEIDD overview: (top) visual decision support system environment; (bottom) detailed decision system highlighting visual recognition and its subsequent
inputs leading to a final decision.

since ceased operations or changed product specifications complex variables, the system relies on three primary, easily
frequently. determinable factors: the product’s outer shell material, its
In this study the fuzzy inference system with desirable weight, and its overall dimensions. Three corresponding data,
predefined outputs has been initially built with empty frame­ widely available for any product released to any market. Such an
work and unknown rules and membership functions. Then approach ensures that the product’s disassembly complexity is
key parameters from e-wastes characteristics for material easily understood.
extraction that could be extracted easily, have been identified It’s essential to emphasise that the selection of these criteria
to systematically sculpt the fuzzy inference system (reverse is not arbitrary. Their universality and straightforward nature
engineered). Extracted criteria: Evaluation, Intactness, cater to the diverse challenges e-waste presents. Whether an
Difficulty, and final Decisioning then integrated into the operator is highly trained or relatively new, the criteria provide
fuzzy inference system and configured to incorporate the pri­ a structured yet simple framework to decide on the optimal
mary VEIDD system. recycling pathway. Subsequent sections will delve deeper into
Beginning with ‘evaluation’, it serves as a practical metric to the intricacies of each criterion. However, it is imperative at
ascertain the inherent recycling potential or value of a product at this stage to underscore the significance of their strategic
the time of supply. Instead of needing intricate manufacturer alignment in enhancing the comprehensiveness and function­
details or product blueprints, this criterion provides a direct ality of the VEIDD system.
gauge of how worthwhile a product is in the recycling context. The VEIDD framework embraces the adaptability of fuzzy
The ‘intactness’ measure further simplifies the assessment by logic, particularly in its decision-making processes following
offering a clear, representation of the product’s current state, visual recognition. In e-waste management, where a myriad of
capturing any wear, damages, or missing components by simply variable factors intersects, deterministic outcomes can be elu­
measuring product weight. The true ingenuity, however, lies in sive. Fuzzy inference systems come to the forefront in such
the ‘difficulty’ assessment. Instead of delving into a myriad of scenarios, interpreting human-rationalised rules over abstract
 INTERNATIONAL JOURNAL OF SUSTAINABLE ENGINEERING 85

mathematical representations. Given the intricate nature of of the pre-trained AlexNet for product recognition, all housed
e-waste end-of-life management, this approach offers signifi­ on the MATLAB platform.
cant advantages in navigating the less-defined mechanisms For the visual recognition setup, we utilised three dis­
and circumstances. tinct cameras, each feeding into the AlexNet on the
A noteworthy advantage of the VEIDD framework, besides MATLAB platform. The topmost Camera 1 captures
its current deterministic pathways for material extraction, is its a bird’s eye view, while the side views are provided by
future-proofing potential. The system is inherently designed to Cameras 2 & 3. Each camera’s output is channelled inde­
be adaptable, making it capable of accommodating emerging pendently into the AlexNet, which has been trained to
technologies, innovative methods, and evolving strategies in recognise and classify products based on varying perspec­
e-waste management as they arise. This ensures that VEIDD tives. This configuration, leveraging multiple cameras input
remains relevant and efficient in addressing both current and and classifications based on single network, adeptly handles
forthcoming challenges in the field. the variations in how a product may be presented, account­
ing for all possible orientations. While our study employed
AlexNet (Deep Learning Toolbox Model for AlexNet
E-waste products’ visual recognition Network ver. 19.1.0) for visual recognition of a small pro­
The AI-based visual recognition stands as a crucial component duct sample set, the system’s architecture isn’t confined to
of the automated SIM concentration and recycling process. a specific machine learning algorithm or recognition
Without such rapid and reliable product recognition, navigat­ approach.
ing the rising volume and diversity of e-wastes would be an Post-classification, the recognition results undergo
insurmountable challenge in automated recycling. Designed to a validation process against an e-waste designated database.
demonstrate VEIDD’s capabilities in a lab environment, our This validation becomes paramount when transitioning to
visual recognition system (as depicted in Figure 6) incorpo­ physical operations, like robotic disassembly. The efficacy of
rates three cameras and utilises a transferred learning version the decision-making process and subsequent actions hinges

Figure 6. (a) Designed visual recognition system embedded with robot operation rig. (b) Actual image of developed visual recognition system while testing
classification of a subject.
86 E. SIMAEI AND S. RAHIMIFARD

crucially on the accurate classification of the e-waste’s type and confidently categorises the product as recognised.
model. It’s worth noting that, while our setup was designed for Conversely, if Prec (top) is 0.95 or lower, the product is marked
laboratory demonstrations, the envisioned real-world applica­ as ‘unknown’. This indicates that the network might require
tion assumes this system would be integrated with a conveyor new or augmented images of the unclassified product, neces­
belt system, swiftly recognising products as they move along. sitating a retraining of the sub-system. In scenarios where
The classification output of our proposed system under­ Prec (top) is between 0.95 and 0.98, the result from the top
goes a validation process to ensure it remains within accep­ camera is cross verified against the classifications from the
table limits. The inherent nature of supervised AI side camera images. If the combined classification scores
classification involves comparing the input image with all from all three cameras surpass 0.95, the product is cate­
trained images across predefined categories and subsequently gorised as recognised. However, if the combined score falls
providing a probability score. This score ranges from 0 to 1.0, below 0.95, the item is set aside for additional imaging and
where 1.0 indicates a perfect match and 0 suggests no potential network retraining. Figure 7 depicts this entire
similarity. process, illustrating the initialisation, classification, and
During validation, if the recognition probability from the result matching stages, using an e-waste sample (for instance,
top camera, termed as Prec (top) is 0.98 or higher, the system a Dell 1520 laptop).

Figure 7. Visual recognition and validation for e-waste sample, dell-1520: (a) Top view, Cam1, 98% likelihood; (b) Side view, Cam2, 96% likelihood; (c) Another side
view, 94% likelihood; average confirmed likelihood 96%.

Figure 8. Workflow of the AI-Driven Product Recognition System in the automated material recovery framework.
 INTERNATIONAL JOURNAL OF SUSTAINABLE ENGINEERING 87

Figure 8 offers a detailed depiction of the product recogni­ behind VEIDD’s visual recognition requires snapshots from var­
tion system integrated within our proposed automated mate­ ious angles of the initial products, it’s the e-waste database that
rial recovery framework. To simulate a real-world setting serves as the primary reservoir of crucial data pertaining to the
where end-of-life products arrive via a specifically designed, products.
electronically controlled conveyor, three strategically posi­ This e-waste database, tailored for VEIDD, holds key
tioned cameras in the lab were set up to mimic this conveyor information regarding the initial value and characteristics
system. These cameras capture instantaneous images from of products. Such details, even though not directly obtained
various angles, ensuring the desired accuracy is achieved, from manufacturers, can be sourced from product data­
thereby laying the groundwork for creating a dataset integral sheets available online or simply gathered through auto­
to image recognition. This strategic arrangement enabled the mated bots or manually entered by operators after a brief
capture of varied images of each of the 12 nominated e-waste inspection. For instance, whilst the weight of a product can
products, which then served as foundational training material be automatically gauged and logged into the database,
for refining the AI network’s recognition capabilities. specific visual features such as external shell material or
Following this setup, actions 3–5 detail the subsequent image certain other data points might necessitate human inter­
preparation stages, encompassing segmentation, cropping, and vention during system training.
enhancement, leading to AI-driven recognition and supervised Despite the current lack of a comprehensive, uniform data­
classification. The AI recognition’s classification outcome oper­ base for e-waste similar to IDIS for end-of-life vehicles (IDIS
ates on a probabilistic basis. If a result or the average of results 2021), we’ve designed a user-friendly database, built using
surpasses a pre-set threshold, it’s accepted (as denoted in action MATLAB software, to suit VEIDD’s unique requirements. As
6 in Figure 8), designating such products as ‘identified products’. shown in Figure 9, this platform facilitates easy data entry and
Conversely, if the classification output is uncertain or below the retrieval. The user can select a product category from
predetermined threshold, the system marks the product as a dropdown list with predefined options and enter specifics
a ‘new entry’. This delineation prompts the need for user (opera­ like the product model, original weight, manufacturing year,
tor) input and further training for the recognition system, as and relative price. In scenarios involving older products with
illustrated in actions 7.1 and 7.2 of Figure 8. inaccessible information and evaluation, estimated values can be
provided. For instance, product valuations might be categorised
as High value for items priced £500 and above, Medium value
E-waste allocated database for those between £50–499, and Low value for items below £50.
For VEIDD to operate effectively as a decision support system for Product age is similarly categorised, and for dimensions, most
de-manufacturing high value and high-volume e-wastes, an e-wastes, being assumed to have a cuboidal form by general
extensive database is indispensable. Whilst the neural network perception, have fields to input length, width, and height values.

Figure 9. Overview of developed front in MATLAB to obtain required data for classification and decision support system to be stored in database. (a) product categories
(c) product initial price (c) age of product (d) near cuboidal size of the product (e) shell material (f) product nominal weight & battery weight.
88 E. SIMAEI AND S. RAHIMIFARD

However, the adoption of legislative frameworks like the evolving consumer market. Such products, with their contem­
European Integrated Product Policy (IPP) directive (European porary tech and components, often carry a significant potential
Commission Environment 2011) has signalled a shift towards for reuse, refurbishment, or remanufacturing. This also impli­
more responsible environmental practices. This directive man­ citly recognises that newer products tend to adopt eco-
dates manufacturers to disclose specific product information friendlier materials, in line with contemporary environmental
before its market release. Such transparency is invaluable, standards, making them prime candidates for sustainable recy­
particularly for e-waste management, as it provides the exact cling. However, this method is not just about identifying the
kind of data essential for populating our VEIDD database. This high value; it’s about casting a net over the vast sea of e-wastes.
ensures a streamlined approach to recycling and underlines The categories, ‘Med-high value’, ‘Low-Med’, and ‘Low value’,
a shared dedication to sustainable e-waste practices within the serve to segment products, providing actionable insights even
industry. in the absence of exact material data.
This approach becomes all the more invaluable when we
consider its adaptability. While designed in the present con­
Evaluation estimation text, it harbours the flexibility to accommodate future devel­
opments in e-waste management, ensuring that the VEIDD
Evaluating a digital product for recycling, especially in the
system remains relevant and effective. Nonetheless, we must
rapidly evolving world of electronics, is undeniably intricate.
acknowledge the method’s natural limitations, such as the
The sheer complexity arises not only from the vast array of
occasional need for user intervention and its lack of pinpoint
components and materials hailing from global supply chains
accuracy in tracking specific material quantities. Yet, in
but also from the ever-changing nature of technological
a domain clouded with unknowns and rapid shifts, this esti­
advancements. This dynamism means that even manufac­
mation method stands out as a beacon, guiding efficient
turers themselves might be hard-pressed to provide
e-waste management.
a comprehensive inventory of every raw material embedded
within their products. Given that some essential details are
often withheld under the banner of ‘trade secrets’ or confiden­
tiality, access to a complete material composition becomes Product intactness analysis
a lofty aspiration for recyclers. To ascertain product intactness, this study draws a comparison
Against this backdrop of uncertainty, the VEIDD system’s between a product’s measured weight and its original nominal
approach – leveraging an indirect evaluation method rooted in weight, serving as an indicator of its potential recoverable
a product’s initial price and its manufacturing year – emerges quality at end-of-life. The nominal weight of each product
as a pragmatic solution. While this might seem like a simplistic can be sourced from its specification list, which is system­
pivot, it holds profound merit. Detailed information, often atically archived in the VEIDD database. As the product
crucial for precise e-waste management, is scarce, making it moves through the SIM concentration process, its weight is
essential to devise strategies that can draw reliable inferences easily determined. With advanced weight measurement tech­
from available data. Figure 10 (considering 2022 as the nologies on the market, particularly those tailored for online
reference year) encapsulates this approach through the retailers such as Amazon, this study omits the intricacies of
Generic Fuzzy evaluation method. creating a new weight measurement system, instead presuming
By assigning categories like ‘High value’ to products man­ the product’s weight is already documented in the database.
ufactured recently with a high initial price, we acknowledge the Often, for safety considerations, batteries within e-waste

Figure 10. Fuzzy evaluation estimation of e-waste products based on their characteristic’s parameters.
 INTERNATIONAL JOURNAL OF SUSTAINABLE ENGINEERING 89

Figure 11. Intactness analysis process of e-waste product based on embedded weight measurement mechanism.

products are extracted prior to processing. Hence, the ‘WR’ or parameters: Shell Material (SM), which identifies the type of
weight ratio reflects the product’s intactness. This ratio is material the e-waste product is crafted from, and Overall
derived from the measured weight over the nominal weight, Density (OD), calculated by dividing the measured weight by
augmented by the battery weight if ascertainable, as detailed in the product’s computed cuboid volume. Essential data for
the following equation: these evaluations are already catalogued in the VEIDD data­
base, ready for the ensuing decision-making process.
Building on the criteria defined by the SM and OD para­
Where: meters, we have established four Fuzzy rules. These rules,
Wm = Measured weight of e-waste product elaborated in Table 1, classify the material recovery process
Wn = Nominal weight of e-waste product (without battery) into three distinct difficulty tiers: Normal, Hard, and
Wb = Battery weight of e-waste product Challenging. For instance, a recovery process is deemed
‘Normal’ if the OD lies between 1.0 and 0.5 and the product’s
Figure 11 delineates the potential values for WR, based on exterior is made of softer materials like Aluminium or plastics.
measured weight. When the intactness ratio (WR) sits within In contrast, a ‘Hard’ label is applied when the OD exceeds 1.0,
1.0 ± 5%, it indicates a product ready for recovery procedures. and the shell isn’t composed of tougher materials like glass or
A WR exceeding 1.05 signals the product’s weight surpasses its ferrous metals. This typically signals that the e-waste product
nominal weight, prompting an inspection for anomalies before possesses a certain level of compactness or intricacy. If the OD
further processing. Conversely, a WR falling below 0.95 hints dips below 0.5 and the shell avoids the inclusion of sturdier
at two potential scenarios: either the product remains intact materials, the recovery process is labelled ‘Challenging’, point­
but lacks its battery, or the WR’s diminished value, unac­ ing to the product’s significant weight or size. Furthermore,
counted for by battery weight removal, suggests the product products with shells primarily made of hard materials, such as
has undetermined missing parts. This may compromise its glass or ferrous metals, are automatically categorised as
suitability for streamlined SIM concentration or recovery. ‘Challenging’ due to the challenges their dense or brittle shells
However, the conclusive course of action is steered by the introduce. Figure 12 provides a comprehensive visual repre­
Fuzzy DSS system, incorporating comprehensive data from sentation of the inputs and potential outputs considered in this
all VEIDD subsystems. difficulty assessment for the recovery process.

Difficulty assessment of recovery process Fuzzy decisioning & chronological order of decision
making
The objective of this section is to employ a methodical quanti­
tative approach to determine the difficulty of the material Figure 13 illustrates the integrated approach of the VEIDD
recovery process. In this evaluation, we lean on two primary system, where results from individual sub-systems –

Table 1. Fuzzy rules for SIM concentration process difficulty assessment.

Fuzzy
Rule predicate output Explanation Product example
Rule one If 1.0>OD>0.5 AND SM is Alum-mild OR Normal The product is not extremely compact or complex Laptops
Plastic-any
Rule two If OD>1.0 AND SM is not Glass or Ferrous Hard Product is compact and complex Mobile phones, tablets
Rule three If OD<0.5 AND SM is not Glass or Ferrous Challenging The product is too heavy or too large PC case, LCD monitor
Rule four If OD is any AND SM is Glass or Ferrous Challenging The product could be compact or complex but has hard or Car Satnav
brittle shell material
90 E. SIMAEI AND S. RAHIMIFARD

Figure 12. Difficulty assessment of recovery process prior to SIM concentration operation.

Figure 13. Chronological order of informed decision making for optimum end of life management of e-waste.

encompassing visual recognition, value estimation, product refinement in decision-making as one progresses through each
intactness, and process difficulty assessment – converge to layer. At its core, the system embraces the philosophy of break­
support the identification of the optimal end-of-life (EoL) ing down complex decisions into more digestible, manageable
treatment for specific e-waste items. These results are assimi­ components. Each layer is tasked with a specific role, effectively
lated within VEIDD’s culminating decision framework. handling distinct subsets of information. The layered approach
The figure further elucidates the layered and sequential also offers an additional advantage: the ability to meticulously
architecture of the Fuzzy decision support mechanism. It fine-tune each level separately, ensuring precision and minimis­
underscores the interplay between data manually input by ing potential errors. Moreover, by partitioning the decision-
users and the output furnished by the VEIDD subsystems, making process, the system can dynamically adapt to changes
ensuring a judicious EoL pathway determination for the in one layer without destabilising the others.
e-waste in question. The initial level leans on direct measure­
ments and observations, the succeeding stage extracts data
from the database post the affirmation of visual recognition,
Case study
while the third and fourth layers are rooted in multi-Fuzzy After the successful development of the necessary hardware and
decision-making, converging to a single output in the fifth software, 12 case study products were chosen to validate the
layer. This intricate framework was sculpted using the practical viability of the proposed VEIDD system and its EoL
MATLAB software package (MATLAB ver. R2019b), comple­ decisions. Data for these products was sourced from manufac­
mented by the Fuzzy Logic Toolbox (ver. 2.5) to streamline the turers’ datasheets. Notably, for mobile phones and laptops,
fuzzification and decision processes. a range of product generations and specifications from recent
The design of a multi-layered Fuzzy system within VEIDD is years were included. Following the construction of decision
not arbitrary but a deliberate attempt to ensure incremental system inputs and outputs, fuzzy rules were defined to
 INTERNATIONAL JOURNAL OF SUSTAINABLE ENGINEERING 91

Table 2. Dynamically constructed fuzzy rules for intelligent recycling with four possible outputs.
No. Rule Predicate Multi-Fuzzy Output
Rule one If (intactness) is (Missing parts) OR (Value) is (Low), regardless of other parameters EoL Operation Path is Shredding
Rule two If (Value) is (high) AND (intactness) is (Intact), regardless of (Difficulty) EoL Operation Path is Manual disassembly for Reuse
(whole or parts)
Rule three If (Value) is (high) AND (intactness) is (Intact) AND (Difficulty) is (Hard) OR (Normal) EoL Operation Path is Robotic disassembly
Rule four If (Value) is (High) AND (intactness) is (Intact) AND (Difficulty) is (challenging) EoL Operation Path is Robotic extraction
Rule five If (Value) is (Med-High) AND (intactness) is (Intact) AND (Difficulty) is (Normal) EoL Operation Path is Robotic disassembly
Rule six If (Value) is (Med-High) AND (intactness) is (Intact) AND (Difficulty) is (Hard) EoL Operation Path is Robotic extraction
Rule seven If (Value) is (Low-Med) AND (intactness) is (Intact) AND (Difficulty) is (Challenging) EoL Operation Path is Shredding
Rule eight If (Value) is (Low) EoL Operation Path is Shredding

characterise the system’s overarching behaviour. Table 2 out­ handle e-wastes with the utmost precision and safety. By fine-
lines eight dynamically constructed rules designed for the sys­ tuning the network to an exceptionally high level of accuracy,
tem to automatically adhere to. Crafting concise yet we ensure that the system can make well-informed decisions
comprehensive rules is essential for the system’s expedient deci­ during end-of-life management processes. This approach
sion-making. These chosen rules encapsulate all potentialities in enhances the VEIDD system’s capability to perform tasks
a deterministic decision framework. It’s worth noting that to with a high degree of accuracy and confidence, particularly
maintain clarity, any outliers are addressed outside the Fuzzy in the context of future robotic operations. The deliberate
system. overfitting is a crucial feature that empowers the system to
Prior to testing the VEIDD, chosen Convolutional Neural excel in its mission of efficient and precise e-waste handling.
Network (CNN) based on AlexNet model trained with When overfitting occurs, an e-waste product that has signifi­
a dataset composed of images from 12 distinct e-waste pro­ cant damage, altered appearances, or modifications may not
ducts. These products were meticulously photographed by our initially be recognised as the same type of e-waste by the system.
setup rig, equipped with three cameras, each capturing the However, by capturing new images from various angles and
products from different angles and under constant lighting retraining the network, it can then be identified as
conditions. This comprehensive dataset was pivotal for train­ a subcategory of the original product, albeit with a different
ing the network to recognise and differentiate unique features appearance. This approach offers adaptability, allowing the sys­
specific to each product. The aim was to empower the CNN to tem to recognise even modified or damaged products. By
identify each e-waste item with high accuracy during the test­ enabling a more sensitive classification, it reduces potential
ing phase of the VEIDD system. errors and ensures comprehensive coverage through multi-
Over the course of this study, seven training iterations were angle imaging. Furthermore, the continuous learning nature of
undertaken with tailored modifications to hone the network the method enhances its reliability over time, making it versatile
for optimal performance. The essence of these training itera­ for diverse e-waste types and promoting a generalisable system.
tions was to adjust the network’s weight, ensuring that given As delineated in Figure 15, the defined fuzzy rules lead to
inputs corresponded accurately with expected outputs. While a visually represented decision tree, mapping all potential
the retrained network could classify all training images accu­ decision routes. When solely accounting for the four possible
rately, its true test was its performance on new, previously system outputs, the resultant decision tree becomes notably
unseen images, a critical step to validate the network’s learning intricate. The four hypothesised outputs (EoL operation paths)
capability. Out of the 194 images from 12 e-waste products under evaluation include: manual disassembly (a standard
stocked in the VEIDD dataset, a random 90% (177 images) solution for SIM concentration), robotic disassembly (details
were utilised for training, and the remaining 10% (17 images) found in Li et al. (2018) regarding semi-destructive robotic
were reserved for post-training validation. disassembly), robotic extraction (assumed akin to robotic dis­
Figure 14 presents the outcomes of the final chosen training assembly but more aggressive and targeted), and whole shred­
iteration. This visualisation captures the progressive enhance­ ding. Given varying automation levels, machinery availability,
ment in accuracy as the training advanced. The various fea­ and precision needs, the current system boasts the flexibility to
tures of this deep learning training progress are showcased in handle an expanded set of future output possibilities. Besides
the Figure 14. It comprises the training accuracy trajectory, these four outputs, three decision system inputs have been
which concluded at 100%, and the training loss, which gradu­ previously discussed. Theoretically, the decision-making
ally minimised to an approximate zero. To validate the train­ occurs in mere moments, and the sequencing or arrangement
ing’s effectiveness, a subset of randomly chosen e-waste of input data bears no tangible impact on the system’s function
product images from the testing set was used, and the results or chosen output. Nevertheless, for consistency’s sake, the
were consistently accurate, underscoring the network’s effec­ outlined VEIDD order remains unchanged.
tiveness. A confusion chart, integral for understanding classi­ Regarding tangible development, in an ideal scenario,
fication performance, revealed no misclassifications as e-waste products would traverse a designated location on or
a evidence of reliability. at a conveyor belt’s end, equipped with cameras and weight-
It’s important to highlight that deliberately overfitting the measuring sensors, thereby enabling real-time recognition.
network serves a strategic purpose. The intentional overfitting Post-recognition and weight assessment, the DSS will
is implemented to reinforce the VEIDD system’s capacity to promptly dictate the best EoL pathway, guiding each product
92 E. SIMAEI AND S. RAHIMIFARD

Figure 14. Detailed result of last stage CNN transfer learning.

 INTERNATIONAL JOURNAL OF SUSTAINABLE ENGINEERING 93

Figure 15. Visualised constructed deterministic decision tree with all developed inputs and considering four outputs for the end-of-life operation path.

Figure 16. VEIDD database (stored information of 12 selected case study products) imported to MATLAB as a table to be accessed and used by VEIDD decision support
system.

accordingly. In our study, spatial constraints necessitated Figure 17 provides a clear visual representation of the step-
a stationary setup for product recognition, weight measure­ by-step results generated by the VEIDD system for the iPad
ment, and decision-making. It was still presupposed that all tablet. During Step 1, the system captures images from varied
four EoL operations were available for system selection. The perspectives, initiating the classification process. The resulting
outcomes from the visual recognition system, with transferred classification underscored a 100% match, identifying the test
learning and a compact sample size, were thoroughly validated sample as an Apple iPad.
without significant anomalies. By Step 3, pertinent data related to the iPad was extracted
The 12 case study products, drawn from a diverse product from the VEIDD database, facilitating the computation of
range, serve to showcase potential subsystem inputs and outputs Overall Density (OD = 1.11) and Weight Ratio (WR = 1.01),
and the final fuzzy decision system. The refined deterministic both imperative for Fuzzy decision-making. Step 4 showcases
decision support system illuminates the potential for monu­ the Fuzzy Evaluation outcome (D1 = 0.6666), derived from the
mental enhancements in e-waste recycling, favouring automa­ iPad’s initial high cost and its manufacturing year, subse­
tion and efficiency over the current labour-intensive approaches quently categorising the EoL product as having a ‘Medium-
in treatment centres. For the VEIDD system, a tailored database High’ value (D2 = 0.6666). Step 5 verified the Fuzzy Intactness
was crafted. This database’s documented information was inte­ result, identifying the product as ‘intact with battery’.
grated into MATLAB, providing accessible data for the VEIDD Meanwhile, Step 6’s Fuzzy Difficulty assessment ascertained
decision support system, as depicted in Figure 16. the processing difficulty as ‘Hard’ (D3 = 0.3937). The
94 E. SIMAEI AND S. RAHIMIFARD

Figure 17. VEIDD sequence diagram showing processes and obtained result for each sub-system testing on an EoL apple iPad tablet.

conclusions from Steps 4–6 were then incorporated as input intactness and the associated complexity of the recovery pro­
parameters for the Final Fuzzy Decision system. This amalga­ cess. With a use of simply constructed database at its core,
mation of prior Fuzzy outputs (D1 – D3) fed into the final VEIDD’s agile framework facilitates informed decision-
Fuzzy decision system, culminating in a result of making, steering each e-waste item to its optimal end-of-life
Df = 0.4194. This recommends ‘Robotic Disassembly’ as the treatment.
most fitting EoL operation (in alignment with the Fuzzy Rule Through 12 diverse e-waste case studies, the VEIDD sys­
of 0.25<Df <0.50 for robotic disassembly), aiming to bolster tem’s capabilities have been robustly tested. These case studies
SIM concentration ahead of material recovery. This Apple illuminated the system’s expansive adaptability and under­
iPad case study’s outcomes effectively aligned with the antici­ scored the potential positive impacts of embedding intelligent
pated deliverables of the VEIDD system, reflecting its founda­ decision-making within recycling processes.
tional parameters and incorporated Fuzzy rules. In a broader industrial context, the VEIDD system holds
promise to transform automated material recovery facilities,
offering rapid product detection and broad-spectrum e-waste
Conclusion & future works
management. Its adaptability ensures that a myriad of high-
E-waste management stands as a pressing global issue, dri­ volume and high variety of e-wastes can be efficiently addressed,
ven by concerns around environmental degradation, the pointing towards a sustainable future in e-waste recycling.
continuous depletion of raw materials for electronics man­ Anticipating future enhancements, we envision the
ufacturing, and the swelling volumes of discarded electronic incorporation of more advanced recognition techniques,
products, compounded by rapid global digitalisation. The like fluoroscopy imaging, to delve deeper into product
significance of advancing e-waste recycling methodologies internal assessment. Moreover, marrying the VEIDD sys­
has never been more critical. Recognising these challenges, tem with possible future more targeted material extraction
the introduction of the VEIDD system represents an inno­ methods stands to offer exponential improvements in
vative leap forward. e-waste recovery rates. In summation, the journey of
Harnessing the power of AI, VEIDD seamlessly integrates VEIDD highlights the intersections of AI, sustainability,
visual recognition to identify e-waste products and collate and real-world applicability in addressing e-waste chal­
tangible attributes such as weight. From here, the system lenges. Future research avenues indicate, particularly
intricately evaluates product-specific features, including around understanding the cost-benefit dynamics of each
 INTERNATIONAL JOURNAL OF SUSTAINABLE ENGINEERING 95

end-of-life operation, ensuring that e-waste management European Commission Environment. 2011. European Commission
becomes a sustainable, efficient, and economically viable Environment: Integrated Product Policy (IPP). European Commission.
process for years to come. https://ec.europa.eu/environment/ipp/index_en.htm.
Eurostat. (2021). Eurostat. Recycling Rate of E-Waste. https://ec.europa.
eu/eurostat/databrowser/view/t2020_rt130/default/bar?lang=en.
Forti, V., C. P. Baldé, R. Kuehr, G. Bel, S. Adrian, M. Brune Drisse, Y.
Disclosure statement Cheng, L. Devia, O. Deubzer, F. Goldizen, J. Gorman, S. Herat, S.
Honda, G. Iattoni, W. Jingwei, L. Jinhui, D.S. Khetriwal, J. Linnell, F.
No potential conflict of interest was reported by the author(s). Magalini, I.C. Nnororm, P. Onianwa, D. Ott, A. Ramola, U. Silva, R.
Stillhart, D. Tillekeratne, V. Van Straalen, M. Wagner, T. Yamamoto,
and X. Zeng. 2020. The Global E-waste Monitor 2020. United Nations
Notes on contributors University (UNU)/United Nations Institute for Training and Research
(UNITAR). https://www.itu.int/en/ITU-D/Environment/Documents/
Ehsan Simaei M. E, PhD, PGDip, Research Associate in Intelligent Toolbox/GEM_2020_def.pdf.
Automation and Robotics. Expertise: Robotics for Sustainable, Space Friedrich, C., A. Lechler, and A. Verl. 2016. The Control Architecture
Robotics, Lab Automation. RoVidiasS - a Robotic Visual Disassembly and Assembly System.
Proceedings of the IEEE International Conference on Industrial
Shahin Rahimifard BSc, MSc, PhD, CEng, FIMechE, FHEA Professor of
Technology, 2016-May, 113–118. https://doi.org/10.1109/ICIT.2016.
Sustainable Engineering. Expertise: Sustainable Manufacturing,
Sustainable Design, Automation in Remanufacturing/Reuse/Recycling, 7474735
Business Models. Grimes, S. M., and D. Maguire. 2020. “Assessment of Priorities in Critical
Material Recovery from Waste Electrical and Electronic Equipment.”
Resources Policy 68 (May): 101658. https://doi.org/10.1016/j.resourpol.
2020.101658.
ORCID Herrmann, C., A. Raatz, M. Mennenga, J. Schmitt, and S. Andrew. 2012.
“Assessment of Automation Potentials for the Disassembly of
Ehsan Simaei http://orcid.org/0000-0002-2403-8084 Automotive Lithium Ion Battery Systems.” Leveraging Technology for
Shahin Rahimifard http://orcid.org/0000-0002-7953-4157 a Sustainable World 149–154. https://doi.org/10.1007/978-3-642-
29069-5_26.
IDIS. (2021). IDIS | the International Dismantling Information System.
Data availability statement https://www.idis2.com/index.php.
Jie, L., M. Barwood, and S. Rahimifard. 2015. An Automated Approach
The dataset and trained CNN network used in this AI-based study are for Disassembly and Recycling of Electric Vehicle Components. 2014
available on GitHub at [https://github.com/Sams5879/VEIDD_system. IEEE International Electric Vehicle Conference, IEVC 2014. https://doi.
git]. For further supporting data related to the decision support system, org/10.1109/IEVC.2014.7056158
please contact the corresponding author, Ehsan Simaei, for access under Jin, X., and G. Tian. 2012. “A New Disassembly Network Graph Model
confidentiality agreements to allow for commercialisation. and Its Automatic Generation.” Applied Mechanics & Materials 195–
196:1083–1088. https://doi.org/10.4028/www.scientific.net/AMM.195-
196.1083.
References Kim, H. J., S. Kernbaum, and G. Seliger. 2009. “Emulation-Based Control
of a Disassembly System for LCD Monitors.” The International Journal
Andersen, M. S. 2007. “An Introductory Note on the Environmental of Advanced Manufacturing Technology 40 (3–4): 383–392. https://doi.
Economics of the Circular Economy.” Sustainability Science 2 (1): org/10.1007/s00170-007-1334-z.
133–140. https://doi.org/10.1007/s11625-006-0013-6. Kinoshita, Y., T. Yamada, S. M. Gupta, A. Ishigaki, and M. Inoue. 2020.
Behdad, S., L. P. Berg, D. Thurston, and J. Vance. 2012. Integrating “Decision Support Model of Environmentally Friendly and
Immersive Computing Technology with Mixed-Integer Nonlinear
Economical Material Strategy for Life Cycle Cost and Recyclable
Programming for Disassembly Sequence Planning Under
Weight.” International Journal of Production Economics
Uncertainty. International Design Engineering Technical Conferences
224 (November 2019): 107545. https://doi.org/10.1016/j.ijpe.2019.
& Computers and Information in Engineering Conference (IDETC/
107545.
CIE), August 2012, Chicago, Illinois, 1–9. https://doi.org/10.1115/
Kopacek, P. 2005. “Semiautomatized Disassembly - Some Examples.” In
DETC2012-70392.
IFAC Proceedings Volumes (IFAC-Papersonline) (Vol. 16, Issue 1).
Behdad, S., L. P. Berg, D. Thurston, and J. Vance. 2014. “Leveraging
IFAC. https://doi.org/10.3182/20050703-6-CZ-1902.01449.
Virtual Reality Experiences with Mixed-Integer Nonlinear
Programming Visualization of Disassembly Sequence Planning Laszlo, R., R. Holonec, R. Copîndean, and F. Dragan. 2019. Sorting
Under Uncertainty.” Journal of Mechanical Design 136 (4): 041005. System for E-Waste Recycling Using Contour Vision Sensors.
https://doi.org/10.1115/1.4026463. Proceedings of 2019 8th International Conference on Modern Power
Bentaha, M. L., O. Battaiä, and A. Dolgui. 2015. “An Exact Solution Systems, MPS 2019, 8–11. https://doi.org/10.1109/MPS.2019.8759739
Approach for Disassembly Line Balancing Problem Under Li, J., M. Barwood, and S. Rahimifard. 2018. “Robotic Disassembly for
Uncertainty of the Task Processing Times.” International Journal of Increased Recovery of Strategically Important Materials from Electrical
Production Research 53 (6): 1807–1818. https://doi.org/10.1080/ Vehicles.” Robotics and Computer-Integrated Manufacturing
00207543.2014.961212. 50 (March 2016): 203–212. https://doi.org/10.1016/j.rcim.2017.09.013 .
Bogue, R. 2019. “Robots in Recycling and Disassembly.” Industrial Robot Liu, Q., Z. Liu, W. Xu, Q. Tang, Z. Zhou, and D. T. Pham. 2019. “Human-
46 (4): 461–466. https://doi.org/10.1108/IR-03-2019-0053. Robot Collaboration in Disassembly for Sustainable Manufacturing.”
de Almeida, S. T., M. Borsato, and C. M. Lie Ugaya. 2017. “Application of International Journal of Production Research 57 (12): 4027–4044.
Exergy-Based Approach for Implementing Design for Reuse: The Case https://doi.org/10.1080/00207543.2019.1578906.
of Microwave Oven.” Journal of Cleaner Production 168:876–892. Mao, W. L., W. C. Chen, C. T. Wang, and Y. H. Lin. 2021. “Recycling
https://doi.org/10.1016/j.jclepro.2017.09.034. Waste Classification Using Optimized Convolutional Neural
Difilippo, N. M., and M. K. Jouaneh. 2019. “Using the Soar Cognitive Network.” Resources, Conservation & Recycling 164 (July 2020):
Architecture to Remove Screws from Different Laptop Models.” IEEE 105132. https://doi.org/10.1016/j.resconrec.2020.105132.
Transactions on Automation Science and Engineering 16 (2): 767–780. Nascimento, J. C. D. 2019. “Decision-making and Fuzzy Temporal Logic.”
https://doi.org/10.1109/TASE.2018.2860945. ArXiv 1–11. https://doi.org/10.48550/arXiv.1901.01970 .
96 E. SIMAEI AND S. RAHIMIFARD

Parajuly, K., K. Habib, C. Cimpan, G. Liu, and H. Wenzel. 2016. “End-Of- Vongbunyong, S., M. Pagnucco, and S. Kara. 2016. “Vision-based execu­
Life Resource Recovery from Emerging Electronic Products – a Case tion monitoring of state transition in disassembly automation.”
Study of Robotic Vacuum Cleaners.” Journal of Cleaner Production International Journal of Automation Technology 10 (5): 708–716.
137:652–666. https://doi.org/10.1016/j.jclepro.2016.07.142. https://doi.org/10.20965/ijat.2016.p0708.
Polat, O., O. Capraz, and A. Gungor. 2018. “Modelling of WEEE Recycling Vongbunyong, S., P. Vongseela, and J. Sreerattana-Aporn. 2017.
Operation Planning Under Uncertainty.” Journal of Cleaner Production “A Process Demonstration Platform for Product Disassembly Skills
180:769–779. https://doi.org/10.1016/j.jclepro.2018.01.187. Transfer.” Procedia CIRP 61 (c): 281–286. https://doi.org/10.1016/j.
Radaschin, A., A. Voda, E. Minca, and A. Filipescu. 2012. “Task Planning procir.2016.11.197.
Algorithm in Hybrid Assembly/Disassembly Process.” In IFAC Yeh, W. C., and S. C. Wei. 2013. Simplified Swarm Optimization in
Proceedings Volumes (IFAC-Papersonline) (Vol. 45, Issue 6 PART 1). Efficient Tool Assignment of Disassembly Sequencing Problem. 2013
IFAC. https://doi.org/10.3182/20120523-3-RO-2023.00336. IEEE Congress on Evolutionary Computation, CEC 2013, 2712–2719.
Saidani, M., B. Yannou, Y. Leroy, and F. Cluzel. 2020. “Dismantling, https://doi.org/10.1109/CEC.2013.6557897
Remanufacturing and Recovering Heavy Vehicles in a Circular Yi, J., B. Yu, L. Du, C. Li, and D. Hu. 2008. “Research on the Selectable
Economy—Technico-economic and Organisational Lessons Learnt Disassembly Strategy of Mechanical Parts Based on the Generalized
from an Industrial Pilot Study.” Resources, Conservation & Recycling CAD Model.” The International Journal of Advanced Manufacturing
156 (November 2018): 104684. https://doi.org/10.1016/j.resconrec. Technology 37 (5–6): 599–604. https://doi.org/10.1007/s00170-007-
2020.104684. 0990-3.
Schwab, K. 2015. The Fourth Industrial Revolution | Foreign Affairs. Zhang, X. F., G. Yu, Z. Y. Hu, C. H. Pei, and G. Q. Ma. 2014. “Parallel
Foreign Affairs. https://www.foreignaffairs.com/articles/2015-12-12/ Disassembly Sequence Planning for Complex Products Based on
fourth-industrial-revolution. Fuzzy-Rough Sets.” The International Journal of Advanced
Shumon, M. R. H., S. Ahmed, and S. Ahmed. 2016. “Fuzzy Analytical Manufacturing Technology 72 (1–4): 231–239. https://doi.org/10.
Hierarchy Process Extent Analysis for Selection of End of Life 1007/s00170-014-5655-4.
Electronic Products Collection System in a Reverse Supply Chain.” Zhu, B., and U. Roy. 2015. “Ontology-Based Disassembly Information
Proceedings of the Institution of Mechanical Engineers, Part B: Journal System for Enhancing Disassembly Planning and Design.” The
of Engineering Manufacture 230 (1): 157–168. https://doi.org/10.1177/ International Journal of Advanced Manufacturing Technology 78 (9–12):
0954405414553980. 1595–1608. https://doi.org/10.1007/s00170-014-6704-8.
Tumkor, S., and G. Senol. 2007. Disassembly Precedence Graph Ziqiang, Z., D. Guohong, W. Zhaoren, and Z. Xiangyan. 2015. “Fuzzy-
Generation. 2007 IEEE International Symposium on Assembly and Clustering Based Cost Modeling of Disassembly Planning for EOL
Manufacturing, 104, 70–75. https://doi.org/10.1109/ISAM.2007.4288451 Products.” The Open Mechanical Engineering Journal 9 (1): 546–551.
Ullah, M., and B. Sarkar. 2020. “Recovery-Channel Selection in a Hybrid https://doi.org/10.2174/1874155X01509010546.
Manufacturing-Remanufacturing Production Model with RFID and Zülch, G., and J. Hrdina. 2010. “Control of Disassembly Systems Based on
Product Quality.” International Journal of Production Economics the Division of Labour by Means of Dynamically Adapting Routing
219 (June 2019): 360–374. https://doi.org/10.1016/j.ijpe.2019.07.017. Plans.” Lecture Notes in Business Information Processing 46
Vongbunyong, S., S. Kara, and M. Pagnucco. 2013. “Basic Behaviour LNBI:396–407. https://doi.org/10.1007/978-3-642-12494-5_35.
Control of the Vision‐Based Cognitive Robotic Disassembly Zülch, G., and R. Schwarz. 2008. “Planning and Balancing of Disassembly
Automation.” Assembly Automation 33 (1): 38–56. https://doi.org/10. Systems.” IFIP International Federation for Information Processing
1108/01445151311294694. 257:49–56. https://doi.org/10.1007/978-0-387-77249-3_6.