Artiﬁcial Intelligence-Powered Decisions Support System for Circular

Economy Business Models

Julius Sechang Mboli

1 a

, Dhavalkumar Thakker

and Jyoti Mishra

Teesside University International Business School, Teesside University, Middlesbrough, TS1 3BX, U.K.

Department of Computer Science, University of Hull, Hull, U.K.

Leeds University Business School, University of Leeds, U.K.

Keywords:

Artiﬁcial Intelligence, Decisions Support System, Circular Business Model, Circular Economy Business

Models, AI, DSS, Circular Supply Chain, AI-Powered Model, IoT.

Abstract:

The circular economy (CE) is preferred to linear economy (LE) as it aims to keep resources in use for as

long as possible, extracting maximum value before recovering and regenerating them. This reduces the need

to extract new raw materials and reduces waste, leading to more sustainable economic growth. Contrarily,

LE also known as a ”take, make, use, dispose” model, is based on resources extraction, products creation, and

waste disposal, which can lead to depletion of resources, environmental degradation and several other hazards.

Several barriers are delaying the switching to CE. Artiﬁcial Intelligence (AI) and emerging technologies can

play signiﬁcant roles in the implementation of CE. In this work, A novel AI-powered model that can serve as

a Decisions Support System (DSS) for CE models is proposed and demonstrated. Product life extension is

created via reuse, repair, remanufacture, recycle and cascade loop. The result of the model outperformed the

LE model. The study demonstrates that technologies can enable smart monitoring, tracking, and analysis of

products to support decision-making (DM). AI-powered sensors and devices can monitor the use of resources

in real-time, allowing for more accurate tracking and reporting of resource use.

1 INTRODUCTION

The ﬁnite nature of our planet and its resources

necessitates responsible and sustainable practices

of production and consumption, as evidenced by

the increasing number of environmentally conscious

consumers, and government policies (Brown and

Wahlers, 1998; Machnik and Kr

olikowska-Tomczak,

2022). Unfortunately, today’s economy is a take-

make-dispose model where raw materials are ex-

tracted for making products and then discarded after

use, hence the term ”Linear Economy” (LE) (WEF,

2014). LE is considered unsuitable for both aquatic

and terrestrial environments, since it also generates

plastic wastes which is projected to outweigh ﬁshes

in the ocean by 2050 (EMF, 2017). The population

of the world is estimated to rise by 392% in a cen-

tury (1950 and 2050). As a result, natural resource

consumption is also growing with a positive correla-

tion to country’s per capita GDP(Nobre and Tavares,

2017). Furthermore, LE is disadvantageous in that it

https://orcid.org/0000-0003-1708-3052

is a system where virgin materials keep entering with

little or no reuse at all, thereby encouraging waste

generation. It has been projected that the world’s con-

sumption of raw materials will double by 2060 (The

E-waste Coalition, 2019). The yearly global Munici-

pal Solid Waste generation currently stands at 1.3 bil-

lion tonnes approximately and it is expected to hit 2.2

billion tonnes per annum before 2025 (Hoornweg and

Bhada-Tata, 2012). This will be a rise of 1.2 kg to

1.42 kg per person per day, with e-waste as a critical

component.

The World Economic Forum (WEF) deﬁnes e-waste

as ”anything with a plug, electric cord or battery

(including electrical and electronic equipment) from

toasters to toothbrushes, smartphones, fridges, lap-

tops and LED televisions that has reached the end of

its life, as well as the components that make up these

end-of-life (EOL) products” (The E-waste Coalition,

2019), and it is also termed as waste electrical and

electronic equipment (WEEE)

In 2016, worrying 44.7 million metric tonnes of

European Commission https://tinyurl.com/4wm82ey2

656

Mboli, J., Thakker, D. and Mishra, J.

Artiﬁcial Intelligence-Powered Decisions Support System for Circular Economy Business Models.

DOI: 10.5220/0011997100003467

In Proceedings of the 25th International Conference on Enterprise Information Systems (ICEIS 2023) - Volume 1, pages 656-666

ISBN: 978-989-758-648-4; ISSN: 2184-4992

 2023 by SCITEPRESS – Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)

e-waste was disposed of globally (Bald

e et al., 2017;

The E-waste Coalition, 2019) which is equivalent to

4,500 times Eiffel tower weight. From this, one per-

son throws away 61kg of e-waste everyday (Bald

et al., 2017) and only 20% recycled. E-waste only

makes up 2% of solid waste streams, yet it is the

fastest growing waste stream and constitute a criti-

cal 70% of the hazardous waste (The E-waste Coali-

tion, 2019). A common smart phone could contain as

high as 60 elements of the scientiﬁc periodic table and

many have high chance of recoverability technically.

The volume of e-waste could top 120 million tonnes

annually by 2050 in the worst-case scenario, based on

United Nations University (UNU) estimate. There are

targets in place to minimise waste generation substan-

tially before 2030 based on the Sustainable Develop-

ments Goals (SDGs) (Forti et al., 2020), such as the

SDG 12, which is ”responsible consumption and pro-

duction”. The target is to ”substantially reduce waste

generation through prevention, reduction, repair, re-

cycling, and reuse”. The e-waste sub-indicator for

SDG 12.5.1 is stated as : (Forti et al., 2020)

(TotalEwaste − Recycled)

(TotalEwaste − Generated)

SDG 12.5.1 for 2019 is 17.7%, which suggests

that more still need to be done to increase this rate.

While business are encouraged to embrace CE prac-

tices for the beneﬁts, there seems to be numerous bar-

riers such as uncertainty surrounding circularity deci-

sions. Emerging technologies such as AI which has

became the buzzword of recent can be employed to

address this this barriers. As part of efforts towards

these goals and more, whether directly or indirectly,

this paper proposes the use of cutting-edge technolo-

gies such as the Internet of Things (IoT), AI, and Ma-

chine Learning in building a Circular Economy Busi-

ness Model (CEBM). This paper particularly focuses

on the use of AI-powered DSS as a circular business

model that aims to help businesses in making data-

driven circularity decisions to achieve resource efﬁ-

ciency, utilisation, productivity, and business bene-

ﬁts. The remainder of the paper is as follows: Sec-

tion 2.1 will explore the barriers and opportunities of

transitioning to CE while Section 3 will present the

AI-Powered DSS model including the results and dis-

cussion, before ending the paper with conclusion and

recommendations in 4

2 TRANSITION TO CIRCULAR

ECONOMY: THE BARRIERS

AND OPPORTUNITIES

As businesses are pressured to pursue the triple bot-

tom line (TBL) which is social, environmental and

economic sustainability, there are several barriers and

opportunities in doing this. In this section, the paper

will ﬁrst identify some of these numerous barriers and

opportunities, before going on to discuss the proposed

solutions.

2.1 Barriers and Challenges to Circular

Economy Implementation

Researchers have identiﬁed that there are various

business beneﬁts of switching to CE. “Prof-

itability/market share/beneﬁt”, “cost reduction”,

and “business principle/concern for environ-

ment/appreciation”, are among the top 3 bene-

ﬁts/drivers of CE (Agyemang et al., 2019). These

and other factors are putting pressure on businesses

to implement CE practices which indeed, some busi-

nesses are willing to switch but are facing challenges.

“Unawareness/uncertainties”, “cost and ﬁnancial

constraint”, and “lack of expertise”, are among the

eminent barriers/challenges in implementing CE

(Agyemang et al., 2019). By grouping the barriers

into TBL framework (Badhotiya et al., 2021), social

barriers include Low demand and acceptance of

remanufactured products, Lack of a standard system

for data collection and performance assessment,

Reluctance to replace EOF products, scepticism to

the quality of refurbished and recycled products,

and Lack of technical and qualiﬁed personnel on

CE (Agyemang et al., 2019). Other social barriers

include lack of design tools for circular business

models and circular products (Agrawal et al., 2021),

and associated risk in transitioning from LE to

CE due to uncertainties and inherent complexities

(Agyemang et al., 2019). The economic barriers

are High upfront investment costs and long-term

economic return (Kumar et al., 2019), lack of funding

and tax incentives (Agrawal et al., 2021), lack

of appropriate partners and complexity in supply

chains (SC) (Benton et al., 2017), need for advanced

technologies and facilities (Kumar et al., 2019), The

environmental barriers include lack of incentives to

promote greener activities, low-tech waste resource

management systems (Cuerva et al., 2014), and

lack of adequate technologies used in land-ﬁlling

and disposal methods (Agrawal et al., 2021) among

others.

Artiﬁcial Intelligence-Powered Decisions Support System for Circular Economy Business Models

657

The list of barriers goes on and concerted and mul-

tidisciplinary efforts are needed to address these chal-

lenges (Agrawal et al., 2021). Most of the barriers

are linked to uncertainty in operational or strategic

DM which cutting-edge technologies can play a big

role in proffering solutions, especially in this AI age.

For instance, the problem of ”Lack of a standard sys-

tem for data collection and performance assessment”

(Kumar et al., 2019) can be addressed by leveraging

IoT and AI for tracking, monitoring, and analysing

products’ performance and usage on real-time. This

will then eliminate the uncertainties regarding product

residual values since real-time data will be collected,

stored, and processed for business and circularity de-

cisions. It could help in deciding incentives for re-

verse logistics purposes and creating transparency for

circular SC. While it is not the intention of this paper

to address all the challenges, the paper proposes solu-

tions to address some of these barriers via the use of

industry 4.0 technologies such as AI, IoT, and ma-

chine learning (Elghaish et al., 2022). speciﬁcally,

these technologies can help in tracking products in

real-time, gathering data, and analysing such data on

a real-time basis which could improve business value,

and achieve resource efﬁciency and a safe environ-

ment. Technology has the capacity to pursue the var-

ious aspects of the TBL, so the next section will look

at how technological potentials are available for en-

abling and implementing CE.

2.2 Opportunities/Enablers of Circular

Economy

Though there are numerous barriers to CE implemen-

tation as identiﬁed in Section 2.1, there are also many

opportunities and drivers for the implementation of

CE, especially where emerging technologies can

serve as enablers of CE practices to overcome the

challenges (Lopes de Sousa Jabbour et al., 2018;

Mboli et al., 2022; Bressanelli et al., 2018). Some

of these emerging technologies will be identiﬁed in

this section. In the construction sector, there have

been efforts that proposed the integration of IoT

and deep learning for detecting the deterioration of

structural health for bridges’ elements caused by

environmental factors. This can extend the lifecycle

of these elements in operations when dedicated

early(Elghaish et al., 2021). Digital technologies

could also be used for innovative business models

that aim to pursue CE practices. For instance, IoT

and Big Data analytics are being used by businesses

to servitise business today (Bressanelli et al., 2018).

As is the case with most academic work in this area,

this work was conceptual, limited to the construction

sector, and lacks real-time data collection which is

critical for CE. The use of technologies as enablers

for CE requires collection of real-time data for

effective analysis and decisions making (Agrawal

et al., 2021). Therefore, a framework that uses

low-cost sensors in reusable products or devices to

gather data was proposed in (Ramadoss et al., 2018).

The framework employs AI to analyse the collected

data so that reusable materials can be detected and

eventually reused for other products. While this

appears like a step in the right direction, it was a

theoretical/conceptual discussion that lack practical

implementation and evaluations. In practice though,

there are companies that are making efforts towards

CE implementations, though slow but still encourag-

ing. such examples include Philips Lighting, Cisco’s

sports shoes, Arup’s circular building, Uber, Airbnb

and many other examples as highlighted in (Nobre

and Tavares, 2017; Uc¸ar et al., 2020). Another work

also identiﬁed the roles of digital technologies in

supporting CE, based on a literature review with 3

case studies which evaluated the relationship between

CE and digital technologies. Business Model Canvas

was employed for integrating R-principles such as

reuse, remanufacture and recycle for that research

(Uc¸ar et al., 2020). Two roles of digital technologies

were identiﬁed as:

• “Digital technologies as an enabler: how digital

technologies can facilitate the development of CE

and improve the collaborations between actors of

its ecosystem?

• Digital technologies as trigger: how digital tech-

nologies can initiate or lead to innovation pro-

cesses or outcomes or associated organisational

routines and mechanisms?” (Uc¸ar et al., 2020)

These works and others not cited here all indicated

that emerging technologies can play important roles

in the implementation of CE whether as an enabler

or a trigger. Since the main issue starts with the

LE which practices a linear SC, the ﬁrst step is to

create a closed-loop supply chain (CLSC) leading to

CSC which is entirely different from the linear SC

by employing appropriate technologies (Mboli et al.,

2022). However, CLSC is not enough as it basically

represents the combination of forward and reverse

logistics such as movement of goods to consumers

and back to the original destination. A more preferred

system is the CSC which is not just about closing the

loop but also considers how circular the system is,

the value it creates, how resource efﬁcient and how

sustainable it is (Mboli et al., 2022; Br

andstr

om and

Eriksson, 2022; de Lima et al., 2022). A CSC entails

a ﬁrm reusing or repurposing products, components,

ICEIS 2023 - 25th International Conference on Enterprise Information Systems

658

or materials and returns from customer to convert

same into new or refurbished products. This can

either be for same usage or different usage altogether

but it aims to minimise the use of raw materials and

waste generation. Due to its advantage over a CLSC,

it can be augured that scholars have lately raised

interest in exploring CSC and its beneﬁts especially

management of uncertainties (de Lima et al., 2022).

Therefore, a novel innovative circular business

model that can serve as a DSS for business is pre-

sented. The model leverages on 4 Rs (reuse, repair, re-

manufacture/refurbish, and recycle) which are mostly

discussed in CSC literature (de Lima et al., 2022;

ResCoM, 2017). The novel model also draws strength

from a cascading framework presented in (Campbell-

Johnston et al., 2020). Hence, the ﬁve main terms that

this AI-powered model builds on are reuse, repair, re-

manufacture, recycle, and cascade. In practice, The

work will use technology to create 5 different cycles

in such a way that a product, its components, or its

inherent materials will pass through based mainly on

it lifecycle and use cycle and the next section will dis-

cuss how this is done.

3 THE AI-Power DSS FOR

CIRCULAR ECONOMY

BUSINESS MODEL

As many studies pointed out some barriers mentioned

in Section 2.1, most of the barriers are either di-

rectly or indirectly linked to technology. For instance,

lack of SC integration and effects of SC complexity,

lack of industrial support, quality of ﬁnished prod-

ucts, proﬁt and market demand level, associated un-

certainty risk, lack of technical and technological ca-

pacity, cost and ﬁnancial constraint, and top manage-

ment resistance were pointed out in (Agyemang et al.,

2019). It can be argued that a robust circular business

model with real-time, tracking, monitoring and prod-

uct analysis will provide the ability for businesses to

make both operational and strategic data-driven deci-

sions. Therefore, the top management will likely em-

brace it once decisions making becomes easier, SC

visibility and transparency could also help in mitigat-

ing the SC complexities. And this could a long way

in answering the challenges. Some businesses are un-

able to reuse materials from returned products since

the status or quality of those is unknown haven been

with users for sometimes. other barriers already iden-

tiﬁed from the analysis of scholars include the lack of

a standard system for data collection and performance

assessment and poor CE knowledge, scepticism to the

quality of refurbished and recycled products, need for

advancement of technology and facilities, and high

cost of establishing eco-industrial chains (Badhotiya

et al., 2021). Following what emerging technologies

are capable of accomplishing as discussed in Sec-

tion 2.2, it can be argued that this is right to explore

the capability of these technologies in domains with

no standard taxonomies with numerous uncertainties.

Hence, this paper proposed an AI-Power DSS for CE

Business Model which will be discussed in the fol-

lowing section.

3.1 AI-Power Decisions Support System

for Circularity Decisions

This is an AI component of a CEBM as proposed in

(Mboli et al., 2022) that utilises the power of AI for

material circularity and business decisions. The work

is a transformation of linear forward SC to a CSC

via enabling technologies. Various AI types such as

expert systems, rule-based systems, machine learn-

ing, deep learning, fuzzy logic systems, and so on

exist, however, this work employs a hybrid AI sys-

tem consisting of rule-based, machine learning, and

fuzzy logic systems. There are basically ﬁve circu-

lar keywords that the model depends on, which cre-

ates ﬁve different routes for the efﬁcient and circular

ﬂow of materials, products, and components. These

ﬁves routes are also referred to as classes in techni-

cal terms for implementation in IoT and AI-enabled

circular model as depicted in 1 and discussed in the

following paragraphs.

The ﬁrst route is Reuse which is a term that covers

all operations where a product is, or its components

are, put back into service for a new use cycle. Com-

ponents of a product can be reused in a new product

as well. (ResCoM, 2017). In the context of this work,

reuse (Figure 1) speciﬁcally refers to functional prod-

ucts/components that the ﬁrst users intend to change

due to other reasons other than failure. It implies the

product is ready for a secondary user without any up-

date or upgrade and can be redistributed almost im-

mediately. The real-time data and analysis provide the

ﬁrm with up-to-date status and value of the product,

lowering the complexity of DM regarding the prod-

uct. This enables businesses to make decisions such

as the right amount of incentives, the logistics costs,

and the cost of securing a secondary market with min-

imal or no risk.

The second route is Repair, which focuses on cor-

recting “speciﬁc faults in a product to bring it back

to satisfactory working condition” (ResCoM, 2017).

Here, the repair route/class will normally apply to

Artiﬁcial Intelligence-Powered Decisions Support System for Circular Economy Business Models

659

Figure 1: An Internet of Things-enabled decision support

system for a CEBM. Source: (Mboli et al., 2022).

products or components that need just minor ﬁxes to

get them working again. The costs associated with

this route will normally be low which the AI model

can predict easily and also support DM around incen-

tives and secondary market.

The third route is Remanufacture which “denotes

the process of disassembly of products into compo-

nents, testing and recombining those components into

a product of at least original performance”. (ResCoM,

2017). In most cases, the resultant new product might

be given a warranty that is similar to that of an equiv-

alent product manufactured out of all new parts but

this will normally be decided by the business. Given

that Refurbishment refers to the process of returning

a product to a satisfactory working condition (Ghar-

falkar et al., 2016; ResCoM, 2017), this work cate-

gorises refurbishment to the same class as remanu-

facturing to avoid the disagreements and ambiguity

between the two terms.

The fourth route is Recycle which is “the pro-

cess that recovers material from products at the end of

their lifecycle. The materials recovered feed back into

the process as feedstock for the original or other pur-

poses” (ResCoM, 2017). The focus here is on upcy-

cling which can be described as “upgrading because

the resulting outcomes still need to enter the recycling

infrastructure” to create a product of higher quality or

value than the original (Korley et al., 2021). The pur-

pose here is to pursue zero waste, a CE key target,

hence, its choice over downcycling. Downcycling is

“recycling materials into new materials of lower per-

formance and/or functionality.” (ResCoM, 2017).

The ﬁfth and the last route is Cascade “Cascade

here refers to the recovery of materials which can no

longer be used by the same company/industry or can-

not be employed for the original purpose but can still

serve another purpose whether in the same company

or a different company. For instance, textile material

can serve different purposes such as clothing, furni-

ture, carpeting”, etc (Mboli et al., 2022; Mishra et al.,

2018).

The main aim of this novel AI-powered CE busi-

ness model is to track and monitor the real-time status

of products and components to make material circu-

larity decisions based on the ﬁve routes/classes ex-

plained above, which currently do not exist. The de-

cision is made based on real-time data on lifecycle,

use cycle, usage pattern, temperature, and other fac-

tors as they contribute to the effects of wear and tear

on products, components, and the inherent materials.

“A lifecycle of a product starts when it is released for

use after it has been (re)manufacture. It ends when it

is disposed of (landﬁll/material recycling) or disman-

tled to harvest/reuse its components. The lifecycle of

some (or all) of the components can continue in new

products when the lifecycle of a product ends. If an

essential amount of components form part of the same

new product, the product lifecycle continues in that

product.” (ResCoM, 2017). This implies that a prod-

uct will still fail and not be suitable for use once its

lifecycle has elapsed though it might not have been in

use from the time it was (re)manufactured. Therefore,

if the lifecycle of a product and its components can

be monitored, the barriers relating to scepticism on

the quality of returned products as identiﬁed in Sec-

tion 2.1 can be overcome. Another factor to monitor

on a real-time basis is the quality of the products re-

garding the use cycle or mean time to failure (MTTF)

cycle (Motovilov and Lutchenko, 2022). The factor

is the information about how many times the prod-

uct or components should be used before it fails due

to wear and tear. Therefore, lifecycle depends on the

duration or lifespan of a product/component which is

drawn from constituting materials, MTFC is the rec-

ommended number of times used and the duration of

usage minus the actual number of times and duration

of usage. This leads to a third factor that needs to

be monitored too and that is consumer behaviours or

consumer usage patterns of a product. This also de-

pends on the number of users per product and the sea-

son as well. With this information, the AI and IoT-

backed model is able to provide insights for making

decisions as will be seen later. This will also motivate

businesses in implementing CE as the uncertainty bar-

rier identiﬁed in (Badhotiya et al., 2021) is overcome.

The key component of the model is performing de-

scriptive, diagnostics, predictive, and prescriptive an-

alytics in CEBM is the Intelligent Material Circularity

Detector (IMCD) which will now be discussed in the

next section.

ICEIS 2023 - 25th International Conference on Enterprise Information Systems

660

3.1.1 Intelligent Material Circularity Detector

The IMCD is an intelligent component that works

with real-time sensor data generated from CEBM.

CEBM is an IoT-enabled circular business model that

uses IoT chips to monitor products on a real-time ba-

sis and transmit data via WIFI. It depends on seman-

tic technologies and an ontological model that works

with reliable 5G networks for real-time products and

components tracking, monitoring, and analysis. The

focus of this paper is to discuss the working and ben-

eﬁts of IMCD, which is the re-engineered building

block of CEBM. However, the full details on how

CEBM work, its architecture, the experimental use

case, the datasets, and more information can be found

in (Mboli et al., 2022). In favour of size and to keep

this work within its limits and scope, the paper will

now focus on IMCD only.

Figure 2: Intelligent Material Circularity Detector.

Data is obtained from smart products with the aid

of IoT and reliable 5G networks and analysed with AI

to support DM on the future ﬂow of products. From

the analysis, ﬁve outcomes referred to as routes or

classes discussed earlier are possible as seen in Fig-

ure 2. Every product is monitored against function-

ality ﬁrst, and if it is functional, re-commerce is pos-

sible without further analysis else, a component-by-

component examination is necessary to detect which

component has failed and that may either lead to re-

pair, remanufacture, recycle, or cascade loop (see Fig-

ure 2). The IMCD is dependent on the principle of

AI that utilises a rule-based system, fuzzy logic, ma-

chine learning, and semantics technology that support

DM despite the uncertainties and ambiguities around

residual values of products. Factors such as mate-

rial lifecycle, product use cycle, product usage dura-

tion, temperature, and current status of the product are

considered for this work. Each product is constantly

and automatically monitored for reusability and fail-

ure analysis by examining its components for func-

tionality with the IMCD. This ensures that the mate-

rial stays longer in use via the 5 different routes of

CEBM, thereby keeping the products, components,

and inherent materials in use longer at the highest

value possible. The next subsection will outline how

these AI rules are developed using machine learning,

fuzzy logic, and semantic web technologies.

3.1.2 Tools and Implementation

The work is implemented with Python, GraphDB, and

Protegee from modeling to data collection, analysis,

and visualisations. Fuzzy logic was employed due

to its capability to handle uncertainties including is-

sues of partial or incomplete data. The procedure for

implementation in protegee using web ontology lan-

guage (OWl) is fully explained in (Mboli et al., 2022).

Python programming was chosen for its wide accept-

ability, compatibility, and high-level nature. Imple-

mentation using fuzzy logic follows standard proce-

dures as recommended in (Wang and Mendel, 1992).

Since the focus is to classify products, components,

and materials based on lifecycle, use cycle, usage du-

ration, temperature, and usage patterns. In python, the

implementation package is called “SciKit-Fuzzy”

which contains default membership functions that can

be 3, 5, 7, or customised functions. This provided the

opportunity to transform the independent variables

(properties of products, components, and materials)

and the dependent variables (classes) 2 to the 7 de-

fault membership functions available in Python. So,

instead of the Very Low, Low, High, Medium, and

Very High, IMCD was implemented with excellent,

good, decent, average, mediocre, poor, and dismal

fuzzy memberships. The classes reuse, repair, reman-

ufacture, recycle, and cascade were developed using

customised membership functions and the tempera-

ture was broken into high, good, and poor with the

customised membership functions as well. An exam-

ple of this is shown in Figure 4 for lifecycle only

though all the properties followed a similar approach.

All the properties are the independent variables except

the class (routes/outcomes) is the dependent variable

since the purpose is to predict the class of each prod-

uct, component or material based on its properties.

Scikit-Fuzzy is a collection of fuzzy logic algorithms

intended for use in the SciPy Stack, written in the Python

computing language. https://tinyurl.com/5n7766hh

Artiﬁcial Intelligence-Powered Decisions Support System for Circular Economy Business Models

661

Figure 3: Initial Membership Functions for Durability.

After fuzziﬁcation, Figure 3 then became Figure

4 as implemented with SciKit-Fuzzy for convenience.

Figure 4: Fuzzy Representation of Use Lifecycle as An-

tecedent (Input).

3.1.3 Rules and Decisions Matrix

The seven default intervals are “dismal”, “poor”,

“mediocre”, “average”, “decent”, “good” and “excel-

lent”. In order to prepare the real-time datasets for

modelling using a fuzzy modelling system, the fol-

lowing steps were adopted from (Wang and Mendel,

1992) for transforming the dependednt and indepen-

dent variables to triangular fuzzy sets since it is easier

to represent in embedded controllers as shown in Fig-

ure 4:

• Divides the input and output spaces of the given

numerical data into fuzzy regions.

• Generates fuzzy rules from the given data.

• Assigns a degree of each of the generated rules

for the purpose of resolving conﬂicts among the

generated rules.

• Creates a combined fuzzy rule base based on both

the generated rules and linguistic rules of human

experts; and,

• Determines a mapping from input space to output

space based on the combined fuzzy rule base us-

ing a defuzzifying procedure.

Several rules were created in line with recommen-

dations from CE and other disciplines’ domain ex-

perts (DEs) as it is an interdisciplinary work (Mboli

et al., 2021). For the scope and limited space, the de-

scription of the semantic web can be found in (Mboli

et al., 2022; Mboli et al., 2021) and the machine learn-

ing models used and the entire rule sets are not in-

cluded here, but below is a sample of the rules is be-

low. This is a simpliﬁed rule that depends on the cur-

rent status of the product but it is just one of the many

rules.

“IF lifecycle is good usecycle is excellent at room

temperature THEN class is reuse”

Rule14 = ctrl.Rule(lifecycle[’good’]

&usecycle[’excellent’],classes[’Reuse’])

The implementation of these rules makes classiﬁ-

cations of products and components possible to sup-

port businesses in DM based on the properties. The

same rule is visualised in Figure 5 revealing the vari-

ous conditions as membership functions and the pos-

sible outcomes when simulated with data in Python.

Figure 5: Visual representation of the input n-dimensional

array for rule14.

3.1.4 Use Case and Description of Datasets

The datasets used in this work consist of two parts.

The ﬁrst part is from a company that manufactures

and markets coffee machines via LE. The datapoints

in this dataset include MTTF cycle, lifecycle cycle,

manufacturing cost, logistics cost, etc. as described

in (Mboli et al., 2022) The other part of the datasets

is real-time sensor data from the IoT-enabled model.

The datapoint from this part includes temperature,

start time, end time, usage duration, and so on as seen

in Figure 6 (Though illegible, but it is only an indi-

cation as the datasets is larger than can be included

visibly here).

While it is difﬁcult to fully describe the datasets

and each datapoint here, this section only provides a

brief overview of it for the beneﬁts of understanding

ICEIS 2023 - 25th International Conference on Enterprise Information Systems

662

Figure 6: A snapshot of the datasets showing days, times,

lifecycle etc. of products and components.

Figure 7: Use Case Overview. More at (Mboli et al., 2022).

the working of the AI-powered CE model. The com-

pany is practicing LE and exploring means of circular

business model for manufacturing and marketing cof-

fee machines. Generally, coffee machines are made

of 6 different materials and 7 different components as

grouped in Figure 7. Figure 7 is only for illustration

and has no link with the company that provided the

datasets for this work. Figure 6 It is difﬁcult to make

sense ofOR comprehend Figure 6 without analysis.

This necessitated the analytical processes to draw in-

sights from the data make data-driven decisions that

are environmentally, economically, and socially sus-

tainable.

3.1.5 Results and Discussions

Insights on the classiﬁcation of products/components

into the 5 classes can support a company’s DM over

which component or product to increase/reduce pro-

duction, therefore enabling just-in-time (JIT) strate-

gies and so on. Considering Figure 8, the majority

of the components are classiﬁed into “Reuse” class,

while a few fall into “Recycle” and “Repair” classes.

“Cascade” and “Remanufacture” classes have the

least number of components based on the datasets.

This kind of information can be useful for companies

that are practicing or intend to implement JIT and can

even support inventory management, strategic and op-

erational DM.

Figure 8: Components Distribution by Class showing Reuse

Route with the Highest Quantity.

Similar insights from the datasets is in Figure 9

where products used by day is presented. The ﬁgure

reveals that of the analysed products, more products

are being used on Tuesdays than the other days with

Sundays being the least. This kind of insights can

inform DM for businesses to determine the days of the

week that their products are being used the most, and

which days they will likely experience high product

failures, issues, or returns. This can support business

DM toward CSC planning, return prediction, and so

on.

Figure 9: Products Usage Variations by Day.

Information on products’ temperature variations

can also suggest why and when these products may

fail as seen in Figure 10. With the ability to drill

down into the data, the temperature at which a prod-

uct is being used can affect how long the product will

last. If it is always used at an adverse temperature

other than the recommended temperature, then there

is a high chance that the product may not last up to its

lifecycle and/or MTTF cycle.

CEBM is capable of predicting the future class

of a product or its components based on real-time

Artiﬁcial Intelligence-Powered Decisions Support System for Circular Economy Business Models

663

Figure 10: Temperature Distribution by classes and Years.

datasets on the properties of the products and current

classes using machine learning algorithms. The sim-

pliﬁed and summarised results of prediction is pre-

sented in Figure 11 for easy visualisation and com-

prehension. A similar prediction result is also possi-

ble using SPARQL Queryy as can be seen in 12 im-

plemented in GraphDB. The presented data included

class, returns on sales (ROS), use cycle, lifecycle and

temperature for each product.

Figure 11: Interactive Products Visualization Using

GraphDB.

Figure 12: Product Classiﬁcation from SPARQL Query.

Acquiring data is important only if value can be

derived from the data, else a waste of resources and

time. Therefore, this AI-Powered model was devel-

oped to help businesses in acquiring and making sense

of their data for material circularity and business intel-

ligence. Four standard types of data analytics which

are descriptive, diagnostic, predictive, and prescrip-

tive analytics as discussed in (Menezes et al., 2019)

are used for this work. Descriptive analytic answers

such questions as what is happening in the business?

Comprehensive, accurate, and live data with effective

visualisations such as Figures 8, 10, 9, 11 and 12

seen in this work can be used to answer such ques-

tions. Diagnostic analytics provide reasons to why

what was discovered in the descriptive analytics stage

happened. The possibility to drill down to the root

causes of what is happening in the business is done at

this stage to further understand the situation. It entails

the ability to eliminate all confounding information

using integrative dashboards, so that the issues iden-

tiﬁed, and the causes become even clearer for sup-

porting informed DM. Predictive analytics offers an-

swers to the question of what will probably happen if

nothing is done. It investigates if business strategies

have remained moderately consistent over time but

now yielding different outcomes. Historical patterns

and datasets from the use case are employed to predict

speciﬁc outcomes using machine learning algorithms

suitable for each dependent variable as was done in

this work where combinations of technologies were

used for various predictive analysis. Most important

is what happens with the predicted results and that

is prescriptive analytics, requiring DM to take action

based on recommended solutions. Prescriptive ana-

lytics suggest what need to be done to mitigate the is-

sues discovered in the ﬁrst 3 stages of the analysis. At

this point, recommended actions based on champion

and challenger testing strategy is employed which ex-

plains why multiple strategies and techniques includ-

ing DEs were used in developing the model (Menezes

et al., 2019). Advanced analytical techniques could

then be applied to make speciﬁc recommendations

regarding products circularity and returns/values as

seen in Figure 13. A comparison of the performance

of the AI-powered model presented in this work and

LE model shows that the novel AI circular model pre-

sented here outperformed the LE in all scenarios (see

Figure 13). The scenarios presented are for reuse,

repair, remanufacture and recycle. Having demon-

strated how technologies can be used to provide DM

support to businesses in this work, the next section

will focus on conclusions, recommendations, and fu-

ture research directions.

Figure 13: Model Contribution and Business value.

The ontological model was evaluated for technical

ICEIS 2023 - 25th International Conference on Enterprise Information Systems

664

compliance in OOP! scanner as presented in (Mboli

et al., 2022) while the DSS was evaluated for quality

of modelling and domain coverage with DEs as pre-

sented in (Mboli et al., 2021).

4 CONCLUSION AND

RECOMMENDATIONS

This paper proposed an AI-powered DSS model that

leverages hybrid AI concepts to enable CE principles.

The results of the novel model outperformed LE as

seen in Figure 13, where reuse scenario produced a

ROS of 60% as against 40% of LE. This is due to the

fact reuse loop has no manufacturing costs but there

is optimised cost for secondary market logistics and

incentives. The work provides businesses with real-

time tracking, monitory and data analytics of their

products which then reduces complexity due to uncer-

tainties and supports DM. With these results, it sug-

gests that emerging technologies can help achieve the

following:

Resource efﬁciency and optimisation: AI algo-

rithms are capable of optimising the use of resources

in manufacturing and production processes, reducing

waste and increasing efﬁciency. As demonstrated,

reuse, repair, remanufacture, recycle and cascade

were created as a means of product life extension en-

abled with technologies.

Predictive maintenance: With the real-time de-

scriptive, diagnostic, predictive, and prescriptive an-

alytics as done in this work, AI can help predict when

an equipment, product or its components is likely to

fail, allowing for proactive maintenance that can ex-

tend the life of these products, components or re-

sources via re-commerce. This supports SDG 12 and

will potentially contribute to the e-waste sub-indicator

for SDG 12.5.1 now at 17% as discussed in Section 1.

Recycling and waste management: The model and

indeed AI can be used to sort and classify waste for re-

cycling, making it more efﬁcient and eliminating the

amount of waste sent to landﬁlls with upcycling.

CSC management: Another critical area of CE is

the circular ﬂow of materials and products in the SC,

reducing waste and making it easier to track the origin

of products/materials which AI can optimally enable.

This goes beyond SC visibility as this model also en-

abled real-time tracking, monitoring, and analysis of

products, components and materials whether in tran-

sits or already in use at the users’ end.

The bottlenecks encountered during this work in-

clude a lack of standard taxonomies for CE, a lack

of evaluation frameworks for interdisciplinary models

such as the smart AI model, a lack of data availabil-

ity, and limited DEs as CE is relatively new. There-

fore, future efforts in this area could be channeled to-

ward addressing these limitations. The particular fo-

cus of this paper was on a business-only DSS where

businesses are able to address uncertainties around

products residual values, incentives, secondary mar-

ket costs and other re-commerce-related costs before

delving into CE. Therefore future research in this area

should consider DSS for other SC partners so that

they are also able to make informed decisions. Other

areas that need to be researched include the explain-

ability and interpretability of complex AI models such

as the one presented here to encourage usability and

acceptance as this was picked up by DEs during eval-

uations. As it is the case with many IoT and AI ap-

plications, Security and privacy issues remain a con-

cern and were also highlighted by DEs and evaluators

of this work. So this is still a challenging area that

future researchers could look into. Policies around

market-based incentives or ﬁnance are other aspects

to consider if CE practices is to be recommended to

all businesses and users.

REFERENCES

Agrawal, R., Wankhede, V. A., Kumar, A., and Luthra, S.

(2021). Analysing the roadblocks of circular economy

adoption in the automobile sector: Reducing waste

and environmental perspectives. Business Strategy

and the Environment, 30(2):1051–1066.

Agyemang, M., Kusi-Sarpong, S., Khan, S. A., Mani, V.,

Rehman, S. T., and Kusi-Sarpong, H. (2019). Drivers

and barriers to circular economy implementation: An

explorative study in pakistan’s automobile industry.

Management Decision.

Badhotiya, G. K., Avikal, S., Soni, G., and Sengar, N.

(2021). Analyzing barriers for the adoption of cir-

cular economy in the manufacturing sector. Interna-

tional Journal of Productivity and Performance Man-

agement.

Bald

e, C. P., Forti, V., Gray, V., Kuehr, R., and Stegmann,

P. (2017). The global e-waste monitor 2017: Quanti-

ties, ﬂows and resources. United Nations University,

International Telecommunication Union, and . . . .

Benton, D., Hazell, J., and Hill, J. (2017). The guide to

the circular economy: capturing value and managing

material risk. Routledge.

andstr

om, J. and Eriksson, O. (2022). How circular is a

value chain? proposing a material efﬁciency metric to

evaluate business models. Journal of Cleaner Produc-

tion, 342:130973.

Bressanelli, G., Adrodegari, F., Perona, M., and Saccani, N.

(2018). Exploring how usage-focused business mod-

els enable circular economy through digital technolo-

gies. Sustainability, 10(3):639.

Artiﬁcial Intelligence-Powered Decisions Support System for Circular Economy Business Models

665

Brown, J. D. and Wahlers, R. G. (1998). The environ-

mentally concerned consumer: An exploratory study.

Journal of Marketing Theory and Practice, 6(2):39–

47.

Campbell-Johnston, K., Vermeulen, W. J., Reike, D., and

Brullot, S. (2020). The circular economy and cascad-

ing: towards a framework. Resources, Conservation

& Recycling: X, 7:100038.

Cuerva, M. C., Triguero-Cano,

A., and C

orcoles, D. (2014).

Drivers of green and non-green innovation: empirical

evidence in low-tech smes. Journal of Cleaner Pro-

duction, 68:104–113.

de Lima, F. A., Seuring, S., and Sauer, P. C. (2022). A sys-

tematic literature review exploring uncertainty man-

agement and sustainability outcomes in circular sup-

ply chains. International Journal of Production Re-

search, 60(19):6013–6046.

Elghaish, F., Matarneh, S. T., Edwards, D. J., Rahimian,

F. P., El-Gohary, H., and Ejohwomu, O. (2022).

Applications of industry 4.0 digital technologies to-

wards a construction circular economy: gap analysis

and conceptual framework. Construction Innovation,

2022(ahead-of-print).

Elghaish, F., Matarneh, S. T., Talebi, S., Abu-Samra, S.,

Salimi, G., and Rausch, C. (2021). Deep learning for

detecting distresses in buildings and pavements: a crit-

ical gap analysis. Construction Innovation.

EMF (2017). The new plastics economy: Rethinking the

future of plastics & catalysing action. https://tinyurl.

com/3rduzf6x. Accessed August 21, 2019.

Forti, V., Balde, C. P., Kuehr, R., and Bel, G. (2020). The

global e-waste monitor 2020: Quantities, ﬂows, and

the circular economy potential.

Gharfalkar, M., Ali, Z., and Hillier, G. (2016). Clarifying

the disagreements on various reuse options: Repair,

recondition, refurbish and remanufacture. Waste Man-

agement & Research, 34(10):995–1005.

Hoornweg, D. and Bhada-Tata, P. (2012). What a waste: a

global review of solid waste management.

Korley, L. T., Epps III, T. H., Helms, B. A., and Ryan, A. J.

(2021). Toward polymer upcycling—adding value and

tackling circularity. Science, 373(6550):66–69.

Kumar, V., Sezersan, I., Garza-Reyes, J. A., Gonzalez,

E. D., and Moh’d Anwer, A.-S. (2019). Circular econ-

omy in the manufacturing sector: beneﬁts, opportuni-

ties and barriers. Management Decision.

Lopes de Sousa Jabbour, A. B., Jabbour, C. J. C., God-

inho Filho, M., and Roubaud, D. (2018). Industry 4.0

and the circular economy: a proposed research agenda

and original roadmap for sustainable operations. An-

nals of Operations Research, 270(1):273–286.

Machnik, A. and Kr

olikowska-Tomczak, A. (2022). Aware-

ness rising of consumers, employees, suppliers, and

governments. In Responsible Consumption and Pro-

duction, pages 22–36. Springer.

Mboli, J. S., Thakker, D., and Mishra, J. L. (2022). An

internet of things-enabled decision support system for

circular economy business model. Software: Practice

and Experience, 52(3):772–787.

Mboli, J. S., Thakker, D., Mishra, J. L., and Sivarajah, S.

(2021). Domain experts and natural language pro-

cessing in the evaluation of circular economy busi-

ness model ontology. In 2021 IEEE 15th Interna-

tional Conference on Semantic Computing (ICSC),

pages 374–379. IEEE.

Menezes, B. C., Kelly, J. D., Leal, A. G., and Le Roux,

G. C. (2019). Predictive, prescriptive and detective

analytics for smart manufacturing in the information

age. IFAC-PapersOnLine, 52(1):568–573.

Mishra, J. L., Hopkinson, P. G., and Tidridge, G. (2018).

Value creation from circular economy-led closed loop

supply chains: a case study of fast-moving consumer

goods. Production Planning & Control, 29(6):509–

521.

Motovilov, A. and Lutchenko, S. (2022). Calculation of

mean time to failure of infocommunication system

wich failure rate is cycle changed. In 2022 Dynamics

of Systems, Mechanisms and Machines (Dynamics),

pages 1–4. IEEE.

Nobre, G. C. and Tavares, E. (2017). Scientiﬁc literature

analysis on big data and internet of things applications

on circular economy: a bibliometric study. Sciento-

metrics, 111(1):463–492.

Ramadoss, T. S., Alam, H., and Seeram, R. (2018). Artiﬁ-

cial intelligence and internet of things enabled circular

economy. The International Journal of Engineering

and Science, 7(9):55–63.

ResCoM (2017). Rescom shared terminology. https://

tinyurl.com/2hzmcvx3. Accessed April 21, 2019.

The E-waste Coalition, W. (2019). A new circular vision

for electronics: Time for a global reboot. World Eco-

nomic Forum.

Uc¸ar, E., Le Dain, M.-A., and Joly, I. (2020). Digital

technologies in circular economy transition: evidence

from case studies. Procedia cirp, 90:133–136.

Wang, L.-X. and Mendel, J. M. (1992). Generating fuzzy

rules by learning from examples. IEEE Transactions

on systems, man, and cybernetics, 22(6):1414–1427.

WEF (2014). Towards the circular economy: Accelerat-

ing the scale-up across global supply chains. https:

//tinyurl.com/nhjsfxva. Accessed May 10, 2019.

ICEIS 2023 - 25th International Conference on Enterprise Information Systems

666