Content uploaded by Xingzhi Wang
Author content
All content in this area was uploaded by Xingzhi Wang on Mar 04, 2021
Content may be subject to copyright.
Contents lists available at ScienceDirect
Journal of Manufacturing Systems
journal homepage: www.elsevier.com/locate/jmansys
Technical Paper
New Paradigm of Data-Driven Smart Customisation through Digital Twin
Xingzhi Wang
a
, Yuchen Wang
a
, Fei Tao
b
, Ang Liu
a
a
School of Mechanical and Manufacturing Engineering, Faculty of Engineering, University of New South Wales, Sydney, NSW, 2052, Australia
b
School of Automation Science and Electrical Engineering, Behang University, Beijing, 100083, China
ARTICLE INFO
Keywords:
Digital twin
Customisation
Smart manufacturing
Personalisation
Product Service system
ABSTRACT
Big data is one of the most important resources for the promotion of smart customisation. With access to data
from multiple sources, manufacturers can provide on-demand and customised products. However, existing re-
search of smart customisation has focused on data generated from the physical world, not virtual models. As
physical data is constrained by what has already occurred, it is limited in the identification of new areas to
improve customer satisfaction. A new technology called digital twin aims to achieve this integration of physical
and virtual entities. Incorporation of digital twin into the paradigm of existing data-driven smart customisation
will make the process more responsive, adaptable and predictive. This paper presents a new framework of data-
driven smart customisation augmented by digital twin. The new framework aims to facilitate improved colla-
boration of all stakeholders in the customisation process. A case study of the elevator industry illustrates the
efficacy of the proposed framework.
Introduction
Customisation is defined as the strategy of supplying different pro-
ducts and services that satisfy diversified customer needs [1]. Advances
in the Internet of Things are driving rapid growth in data generated by
physical manufacturing systems. This exponential increase in data
source, coupled with data processing techniques, is shifting customi-
sation into an increasingly data-driven paradigm [2]. Hence, traditional
mass customisation is shifting to a more intelligent, autonomous and
rational decision-making paradigm [2,3], known as smart customisa-
tion [4,5]. Despite this increase in data utilisation, existing sources of
big data remain insufficient in fulfilling personalised requirements for
three main reasons. Firstly, present-day product development is sig-
nificantly affected by the availability of real-world data. Insufficient
data inhibits the accuracy of the customisation process. Secondly, ex-
isting insights of customer requirements are passive, superficial and
fragmented, and do not provide the detail needed for higher-level
customisation. Defining more explicit and accurate insights is critical,
given the uncertainty of today’s market [6]. Finally, the current practice
of customisation can only fulfil customer requirements through physical
products. Manufacturers are having difficulties in supporting custo-
mised services when products are with customers [7]. Therefore, smart
customisation solely driven by real-world data is ineffective for the
identification and fulfilment of customer requirements.
From a review of existing literature, the main technical challenge
for facilitating customisation is to satisfy three requirements by a single
approach. Firstly, the current data-driven customisation approach
should be augmented by virtual models to compensate for the absence
of real-world data. Secondly, the data analysis of smart customisation
must become more proactive and user-centricto reduce the uncertainty
of customer demands. Finally, customised product-services should be
offered through the virtual modelling of user context to increase cus-
tomer cohesion [8].
As an emerging technology, digital twin is increasingly applied in
design, manufacturing, and service industries [9]. It aims at creating a
digital representation that mirrors the functionality, structure, and
performance of a physical entity (e.g., a product, a system, or an asset).
It is characterised as a self-reinforcing mechanism that achieves real-
time data transmission between the physical and digital entities. Spe-
cifically, it enables the coevolution between physical and digital entities
by updating the virtual realm with physical data and feeding the im-
provements in the virtual realm into the physical realm [10]. Many
previous studies demonstrate that digital twin is a promising tech-
nology to support product and service development. The integration of
physical and digital entities via digital twin is powerful in addressing
major design challenges.
Given the potential of the digital twin in design, the authors are
motivated to employ digital twin to augment data-driven smart custo-
misation in the context of intelligent manufacturing. The following
paper aims to demonstrate this concept by designing a new framework.
Section 2 reviews some previous studies on customisation in terms of its
developments, and impact on design and manufacturing. Section 3 re-
views previous studies on digital twin concerning its theoretical in-
vestigations, practical applications and enabling technologies. Section 4
introduces the new paradigm of digital twin-augmented data-driven
smart customisation framework, followed by an illustrative case study
in Section 5 to validate its efficacy. Conclusions and future works are
presented in Section 6.
https://doi.org/10.1016/j.jmsy.2020.07.023
Received 4 November 2019; Received in revised form 25 April 2020; Accepted 30 July 2020
Journal of Manufacturing Systems xxx (xxxx) xxx–xxx
0278-6125/ © 2020 The Society of Manufacturing Engineers. Published by Elsevier Ltd. All rights reserved.
Please cite this article as: Xingzhi Wang, et al., Journal of Manufacturing Systems, https://doi.org/10.1016/j.jmsy.2020.07.023
Previous Studies on Customisation
Mass Customisation
Mass customisation refers to the delivering of products and services
for the satisfaction of specific customer needs, yet with near-mass
production efficiency [11]. The emergence of mass customisation
paradigm is driven by society. Increasingly deepened globalisation has
significantly expanded the production capacity of this world, bringing
about an over-supplied market [12]. In this circumstance, manu-
facturers should fulfil the specific needs of customers precisely and
deliver the customised products timely, rather than just offering cus-
tomers with standardised products. Requirement fulfilment and re-
sponsiveness are two essential criteria for success in the customisation
business [13]. Several strategies are required to enable manufacturers
to differentiate products and adapt to demands [14].
The first enabler is modularised design. Modularity involvesone-to-
one mappings between functional requirements and physical product
features [15]. The decoupled interfaces between components allow
each module to be changed independently. As a result, various modules
are introduced separately to satisfy customer’s diverse needs. A good
example of the modular design is Dell laptops, whereby customers pick
the processor, memory, storage and other components for their own
computers from configuration menus.
The second enabler is the agile supply chain. In mass customisation,
product differentiation is usually postponed to the latest possible time
before delivery, which requires an agile supply chain that can flexibly
and quickly deal with “unique” products with unpredictable demands
[16]. To save inventory costs, the appropriate differentiation time point
(decoupling time point) should be carefully considered by manu-
facturers.
The third enabler is advanced manufacturing systems. The in-
dustrial transformation from mass production to mass customisation
contributes to the evolution of manufacturing systems into re-
configurable manufacturing systems (RMS) and flexible manufacturing
systems (FMS) [17]. FMS allows production flexibility with respect to
the type and quantity of parts, which is achieved by modular manu-
facturing processes [18]. RMS is targeted at providing precise func-
tionality and production capacity in real-time through making adjust-
ments according to market requirements [19].
Emerging Paradigm of Data-driven Smart Customisation
In the big data era, data are available everywhere. The advances in
new ITs (e.g., industrial Internet of Things, cloud computing, and data
analytics) and manufacturing systems (e.g., robotics, human-machine
interfaces, and industrial information systems) have substantially en-
hanced the manufacturer’s capability to make more informed decisions,
which will in turn pioneer the new paradigm of data-driven smart
manufacturing, as well as the new paradigm of smart customisation
[2,3]. The smart customisation refers to the customisation paradigm
that incorporates big data to better understand customer requirements,
improve design efficiency and productivity. Notably, manufacturers can
develop customisation strategies regarding specific aspects of the whole
product life cycle. For example, in the design stage, the increasing
availability of data from customers can lead to better identification of
customer requirements. In the manufacturing stage, the connected
shopfloor enables more effective supply chain coordination. Besides,
advanced data analytics can improve the production efficiency by
aiding manufacturers in predicting quality defects, learning order pat-
terns, and adjusting manufacturing systems [20,21,22]. In the usage
stage, maintenance services can be recommended based on the real-
time operating data collected by equipped sensors.
Comparison between Mass Customisation and Smart Customisation
As shown in Table 1 below, three characteristics differentiate smart
customisation from mass customisation.
Firstly, smart customisation is a data-driven approach, whereas
mass customisation is primarily driven by the designer’s domain
knowledge [2,23]. Currently, manufacturers play a dominant role in
mass customisation, meaning that they are responsible for conducting
market research, developing products, and maintaining configuration
menus. However, knowledge-driven customisation is time-consuming
and costly, and its effectiveness is greatly affected by the designer’s
experience level [24]. With the advances of the Internet of Things (IoT),
information technologies, and data analytics, it is possible to use big
data directly to drive various customisation activities. For instance, it is
possible to develop new products and marketing strategies with re-
ference to customer data on the website (e.g., e-commerce platforms
and social networking platforms) [25]. The data-driven approach not
only results in a faster definition of customer preferences and market
trends, but also renders the decision-making process more efficient and
representative.
Secondly, smart customisation is more predictive and proactive.
Increasing availability of data allows manufacturers to better forecast
customer requirements and demand fluctuations [26], and also leads to
a more collaborative supply chain. In this case manufacturers can ac-
tively adjust manufacturing systems and production plans, capture
emerging customer requirements, and seize market opportunities at the
earliest opportunity.
Thirdly, smart customisation facilitates mutual monitoring between
customers and manufacturers, and closely engages customers in the
whole customisation process. Customer participation has shown to be a
promising way of building mutual trust between manufacturers and
customers, and it also increases the level of customer engagement in
customisation activities. Besides, constant monitoring of customer
usage data enables manufacturers to develop a deeper understanding of
customer behaviours and preferences, thus leading to the formation of a
value-added product-service ecosystem.
Previous Studies on the Digital Twin
Theoretical Investigations of the Digital Twin
NASA defines digital twin as an integrated system that simulates a
multi-physical, multi-scale, probabilistic model of a complex product
that uses the best available physical models and sensors to reflect the
condition of its corresponding twin [27]. According to this definition, a
typical digital model has three dimensions, i.e., the physical product,
the virtual model and the connection between the product and the
model. Later on, with the continuous expansion of application fields,
Tao et al. extended the digital twin into five dimensions (physical en-
tity, virtual entity, service, data and connection) [28]. This extension is
Table 1
Comparison of mass customisation and smart customization.
Mass customisation Smart customisation
Driving force of the decision-making process Designers’ experience and domain knowledge Big data
Response to the demand uncertainty Passive, imposing pressure to suppliers Predictable and proactive, enhancing coordination
Relationship with customers Limited customisable options, limited customer participation User interaction, open innovation
X. Wang, et al. Journal of Manufacturing Systems xxx (xxxx) xxx–xxx
2
driven by increasing service demands from different industries and
users, as well as a growing concern for the integration and interaction
of cyber-physical systems [29]. Although academia and industry hold
different views on digital twin, they all regard it promising in mon-
itoring the ongoing status of physical entities, recognising system
complexities, detecting abnormal situations and predicting future
trends.
Research has indicated that four core components constitute the
digital twin [30]. The first core component is model. A digital compa-
nion is composed of a set of models that allow the accurate re-
presentation and 3D visualisation of physical objects on the computer.
Data, which enables the digital twin to mine useful information con-
tinuously and to interact with the virtual world, is the second core
component. Data of the digital twin comes from both physical entities
and the virtual world, including product lifecycle data from the phy-
sical world, simulation data from the virtual world, operation data from
information systems, etc. The third core component is connection.
Seamless connection is critical to the interaction between the physical
world and the virtual world. Connection lays the foundation for the
exchange, integration and fusion of data, and enables a better colla-
boration among elements to achieve iterative optimisations. Last but
not least, service is an equally important component. Aimed at facil-
itating the convenient use of the digital twin, services can greatly en-
hance the practical values of the digital twin, allowing users without
professional knowledge to obtain expected results.
Enabling Technologies of the Digital Twin
Fig. 1 summarises the enabling technologies of the digital twin,
which are classified based on the five-dimensional model [31]. The
physical world is deemed as the prerequisite of effective digital twin
development, but it is usually a complex system with different entities
intricately connected. In this circumstance, knowledge, sensing and
measurement technologies, including static parameters, dynamic be-
haviours, real-time evolution, expert knowledge, etc., should be
adopted to fully understand and perceive the physical world.
Based on understandings of the physical world, various modelling,
simulation and visualisation technologies are adopted to create the
virtual counterpart. Big data imparts intelligence to the built digital
twin. In order to expand data sources, several data mining technologies
are employed to enable the extensive and diverse dataset collection.
Besides, advanced data analytics and fusion technologies function to
convert raw data into useful information for decision making
[32,33,34]. The service module makes the digital twin convenient to
use by encapsulating different functions into standardised application
services (monitoring, simulation, diagnosis and prognosis) using ap-
plication software, platform architecture, service-oriented architecture
(SoA) and knowledge technologies. Finally, several connection tech-
nologies facilitate the seamless interaction of all dimensions in the di-
gital twin. The Internet technologies, for instance, are necessary to
connect physical entities with other digital twin dimensions. Commu-
nication, interface and interaction technologies enable the data ex-
change within the digital twin. Besides, several interactions and cyber-
security technologies are also required in the connection module of the
digital twin.
Applications of Digital Twin
In recent years, many exemplary digital twin applications have at-
tracted attention from both industry and academia. The digital twin can
be applied to different stages of a product’s lifecycle in the context of
manufacturing, including product design, production, and service
stages [9]. In the design stage, the digital twin not only allows designers
to better understand customer requirements and sparks new design
concepts, but also assists designers with the detailed design of products
so that they can more effectivelly refine the design schemes like func-
tions, appearance, performance, manufacturability, etc. Through vir-
tual verification, the digital twin uncovers and resolves all kinds of
product defects virtually, hence significantly reducing the time of ap-
proaching the market. In the production stage, the digital twin carries
out the optimal production planning using data from the physical
shopfloor. In addition, it also functions to monitor and regulate the
manufacturing process to avoid quality risks. In the usage stage, the
maintenance service driven by the digital twin enables service
Fig. 1. Enabling technologies of the digital twin.
X. Wang, et al. Journal of Manufacturing Systems xxx (xxxx) xxx–xxx
3
providers to monitor product conditions and ensure reliability. Another
promising service rendered by the digital twin is Remote Prognostics
and Health Management (PHM), by which, potential disturbances,
faults and defects of sophisticated equipment can be diagnosed re-
motely [35].
Aside from the applications in product lifecycle management, the
digital twin is also applicable to many other industries such as energy,
aerospace, urban planning, civil engineering and healthcare. For ex-
ample, digital twin models can predict potential malfunctions of air-
crafts by mirroring the physical entity and extracting real-time data, so
as to give engineers sufficient time to develop effective solutions. In
addition, the digital twin can be used in the biomedical industry. It can
model and monitor the organ functions with the help of implanted
biotech sensors, thereby enabling the virtual analysis of heart condi-
tions before inserting pacemakers and other devices. What’s more, the
digital twin can support urban planners to improve residents’ quality of
life. Administrators can build a large-scale simulation model and de-
velop long-term planning strategies based on the real-time reflection of
population density, commute condition and infrastructure utilisation.
Digital Twin Augmented Smart Customisation Framework
Applicability of the Digital Twin for Smart Customisation
Based on the above literature, it is foreseeable that the mainstream
of customisation is to adapt the product to customers’ new needs by
predicting customer requirements, enhancing customer participation,
and offering interactive services. However, the current customisation
frameworks are unable to achieve this goal since they only deal with
data concerning the physical world. As a virtual-reality integration
technology, the digital twin leverages sufficient data from both physical
and virtual spaces, and can compensate for the absence of real-world
data via virtual models. Hence, it is beneficial to integrate digital twin
with existing paradigm to enhance customisation capabilities.
As shown in Fig. 2 the current smart customisation paradigm is
primarily driven by real-world data. Real-time data collected, processed
and analysed by industrial IoT and CPS technologies allow designers to
better understand customer requirements and organise customisation
activities. However, without any data fusion, modelling and simulation
technologies, manufacturers’ understanding of those data tend to be
very superficial and fragmented. In this circumstance, the digital twin
can interpret more insights and patterns from data, which will sig-
nificantly enhance the intelligence of customisation. For instance, de-
signers can acquire a deeper understanding of the product context by
integrating user interaction data with surrounding environment data,
thereby promoting the logical reasoning of customer behaviours and
choice decisions. Besides, services offered by the digital twin will sig-
nificantly reduce the difficulty of analysing data and customising pro-
ducts. Reduced knowledge threshold will allow more customers to
better participate in various customisation activities. As a result, cus-
tomisation business may shift towards an open-architecture paradigm.
The digital twin-augmented paradigm of smart customisation has three
key features.
Firstly, the digital twin expands data sources into the virtual space,
which will significantly enhance customisation capabilities in terms of
quality and reliability. To better satisfy customer needs, many compa-
nies have developed thousands of or even millions of possible product
configurations. As manufacturers cannot test all the component var-
iants, serious component integration issues of the customised products
arise. The digital twin, which represents high-fidelity physical entities
and properties via simulations, visualisations and other advanced
computer-aided technologies (CAX), offers great help in resolving these
issues. Thus, manufacturers can effectively handle a much wider variety
of configurations, derive better customisation alternatives and ensure
reliability without the need for testing physical models one by one.
Secondly, the digital twin allows customers to better participate in
the customisation process, especially in the virtual environment.
Traditionally, customisation activities are primarily managed and co-
ordinated by manufacturers, which are responsible for informing sup-
pliers of customisation capabilities, configuration menus, and selection
ranges, as well as coordinating customised components with suppliers.
Driven by the digital twin, the customisation process can be shifted into
an open-architecture platform, through which, self-service customisa-
tion can be achieved. Specifically, it will automate the virtual modelling
and optimisation processes, allowing customers to design and verify
their products even if they only have limited domain knowledge.
Hence, customers can modify their products at any time when it is
necessary, or upgrade the components to adapt to their new needs
Fig. 2. Mapping between smart customisation and digital twin capabilities.
X. Wang, et al. Journal of Manufacturing Systems xxx (xxxx) xxx–xxx
4
through independent collaboration with certified suppliers.
Finally, services offered by the digital twin will become more per-
sonalised. In recent years, service customisation has become a new
competing area for manufacturers, but the main challenge is the mon-
itoring and understanding of customer behaviours and choice decisions.
In this context, the digital twin has profound technical implication for
the service customisation. The digital twin provides a realistic re-
presentation of the user environment so that many unknown using
scenarios are made visible (e.g., heat environment, noise, energy con-
sumption) by manufacturers, which effectively helps manufacturers
better serve their customers.
Data Components of Smart Customisation
Empowered by IoT, cloud computing and data analytics, the pro-
posed framework is driven by big data. Big data in manufacturing in-
dustry refers to a huge volume of data obtained from the whole product
lifecycle in multi-sourced ways [3]. It is characterised by 5 Vs, in-
cluding: volume (huge volume of data), variety (multiple formats of
data, i.e., digital signals, text data, graphic data), velocity (high data
reproduction speed), value (high value hidden behind data) and vera-
city (high reliability and fidelity). The digital twin functions to in-
tegrate big data, upload data into virtual space and extract useful in-
formation from it. All the data generated by customisation activities in
each stage of the product lifespan (design, procurement, manufacturing
and usage) are collected. The data required for the proposed framework
is summarised in Table 2 below.
Design data: Driven by new ITs, a variety of data are available to
facilitate product design. Through the systematic fusion of customer-
interaction data with surrounding environment data, a comprehensive
model of user-profiles and using-context is built. This model helps
manufacturers better understand implicit customer needs. Furthermore,
manufacturers can use online data such as customer reviews, old-gen-
eration product data and competing product data to build product
competitiveness model, thus better predicting market trends, existing
rivalry and emerging customer requirements.
Procurement data: Several procurement-related data are collected
from both physical manufacturing systems and industrial information
systems. Demand data, such as economic conditions, raw material
prices, downstream inventories, are integrated to forecast demand
fluctuations. Supplier’s data are collected and integrated to build the
upstream supply chain model, which helps to prevent potential risks
such as late delivery and component defects. As a result, manufacturers
can effectively streamline the supply chain to ensure on-time delivery.
Manufacturing data: The digital twin enables manufacturers to
monitor and optimise the manufacturing process continuously. The
shop floor data and product data are collected and transmitted to
virtual space where several simulation models function to optimise
production activities. Besides, manufacturers can optimise the machine
setup before switching production lines with the help of the digital
twin, thus greatly reducing the time spent on switching production lines
for customised products.
Usage data: The inclusion of data in the usage stage (such as product
transportation, installation, repairing, and maintenance) will sig-
nificantly expand the customisation capabilities of services. In this
stage, usage data are integrated with product operation data, or sur-
rounding environment data, to explore using habits of customers. By
analysing customer behaviours, big data can assist with the precise
marketing of services, which will significantly improve customer ex-
perience.
Integrated Framework
The integrated framework is shown in Fig. 3. The smart customi-
sation module involves manufacturing systems, customised product and
user environments. Manufacturing systems and user environments are
interacting with the customised product throughout the product life-
cycle from the design stage to the usage stage. The IoT technology
enables manufacturing systems to continuously monitor the production
process so that production plans and parameters can be adjusted timely
to maximise equipment efficiency. Besides, the customised product can
adjust to changing customer behaviours by interacting with user en-
vironments. By monitoring the natural environment (e.g., temperature,
humidity, wind speed, weather, etc.), social environment (e.g., social
relationships, culture, etc.), product operating conditions (e.g., using
frequency, expected usage life, product reliability, etc.) and user con-
texts (user profiles, user behaviours, etc.), products can upgrade
themselves to fulfil dynamic customer needs. Real-time data collected
from user environments can also help designers explore more oppor-
tunities for offering personalised functions or services. In the smart
customisation module, tremendous data generated from the physical
product, manufacturing systems and physical environments will be
uploaded to the virtual space to support the construction of the digital
twin module.
The cloud data analytics mainly converts data from physical entities
into discrete digital signals for the digital twin. It continuously in-
tegrates data in the virtual space to produce useful information for
physical entities. The cloud data analytics has four functions, namely,
data storage, data processing, data analysis and data transmission.
Cloud services allow data to be securely stored and shared in the cy-
berspace and also guarantee the effective classification and manage-
ment of data from the distributed environment. For instance, object-
based storage architecture will allow unstructured data to be freely
integrated when necessary. In data processing, data collected in multi-
Table 2
Big data in smart customisation
Stage Data category Examples
Design Customer data User identity, usage contexts, surrounding environments, purchase information, customer reviews, user interaction, etc.
Design data Concepts, product specifications, configurations, prototypes, functions, models, simulations, tests, etc.
Knowledge data Historical products, model banks, domain knowledge, design rules, ethics, etc.
Procurement Supplier’s data Progress, certification, manufacturing resources, locations, prices, etc.
Component data Inventories, quality, logistics, prices, component availability, etc.
Demand data Market, economic data, political environments, downstream inventories, etc.
Manufacturing Shopfloor data Resource availability, production plans, production capacities, inventories, equipment health, etc.
Product data Attributes, performance, parameters, process conditions, etc.
Usage Operation data Product health, component status, operation environments, etc.
Usage data Use frequency, habits of use, operation conditions, using environments, user profiles, etc.
Service data Installation, repair, maintenance, disposal, transportation, etc.
X. Wang, et al. Journal of Manufacturing Systems xxx (xxxx) xxx–xxx
5
sourced ways are integrated and fused to assist with decision making.
Through a series of data analysis processes, data is converted into
comprehensive information that dynamically model, predict, estimate
and analyse the changes of the physical world. Besides, the patterns and
knowledge derived from data analysis will also be appropriately stored
for further reference. Finally, cloud services act as a bridge enabling
effective data transmission between virtual and physical space.
As the driving force to facilitate the intelligence of the smart cus-
tomisation module, the digital twin module functions to uncover im-
plicit customer requirements, improve product quality and enhance
customer involvements. There are three digital twins involved in the
smart customisation processes, including the product digital twin, the
manufacturing system digital twin and the user environment digital
twin. The product digital twin is the high-fidelity replica of the custo-
mised product, reflecting the entire lifecycle as well as each customi-
sation stage of the product. By continuously analysing and simulating
real-life data, the virtual model predicts and prevents potential cus-
tomer dissatisfaction while exploring hidden customer requirements.
The manufacturing system digital twin truthfully reflects the operation
conditions and resource utilisation of shopfloor and the supply chain,
helping manufacturers identify available resources, optimise produc-
tion processes and enhance supply chain coordination autonomously.
For instance, the manufacturing system digital twin can predict po-
tential bottlenecks caused by changeover, downtime and setup time by
monitoring machine conditions and production plans in real-time, and
then learn from bottleneck patterns and reschedule production re-
sources autonomously. The user environment digital twin represents
the exterior environment where the product is installed and operates. It
can reflect how the product interacts with the environment and users.
The digital twin-augmented smart customisation has several applica-
tion prospects in four customisation stages, as shown in Fig. 4.
Design stage
In the product design stage, data generated from the digital twin
facilitate the holistic exploration of customer needs, and hence increase
configuration effectiveness and reduce customer dissatisfaction in ad-
vance. The digital twin uploads product data in the virtual space to
simulate expected using scenarios, through which designers can quickly
identify implicit demands and resolve potential risks in advance. Rapid
configurators recommend and optimise configurations according to
customer requirements, module functionality, resource availability, etc.
Virtual verification enables designers to test product functionality and
resolve potential product defects in the early stage, thereby sig-
nificantly reducing lead-time. In this stage, several applications are
carried out to facilitate design activities. Firstly, the context where the
product operates is modelled using data from both physical space and
virtual space through the digital twin. Designers can correctly interpret
precise contexts in the virtual space and pre-identify customer re-
quirements, quality metrics or design parameters ahead based on the
product interaction data accumulated in surrounding environments.
After user context modelling, rapid configurators will recommend sui-
table components to designers. By mapping customer requirements into
structured product specifications, big data analytics searches and mat-
ches the most suitable product modules according to their functional
characteristics. The configuration can be optimised to cope with cus-
tomers' unique using contexts. Besides, the product can be verified in
the virtual space to avoid conflicts between components or with en-
vironments.
Procurement stage
In the procurement stage, the digital twin enables manufacturers to
quickly adjust their procurement strategies by analysing demand fluc-
tuations, customer activities, inventory levels, material supply, and
even market trends. Being informed of the real-time conditions of the
Fig. 3. Data-driven smart customisation framework based on the digital twin.
X. Wang, et al. Journal of Manufacturing Systems xxx (xxxx) xxx–xxx
6
supply chain, such as unforeseeable disturbances, irregular orders, and
changing customer needs, manufacturers can adjust order priorities,
simulate alternative procurement strategies, and verify their feasibility.
Applications in this stage can be illustrated from the following aspects:
1) Smart purchase management and 2) dynamic logistics scheduling.
Nowadays, the delivery date of the customised product is de-
termined at the time of signing the contract, which poses a big chal-
lenge of uncontrollable procurement lead time. However, the proposed
framework renders the purchase more flexible to the changes, and
significantly reduces the schedule uncertainties. The production pro-
cesses of customised components (including fabrication, inventories
and logistics) are monitored. Real-time data generated from suppliers
are also monitored simultaneously in the virtual space to accurately
predict arrival time. Increasing certainty in purchase helps manu-
facturers to reduce the safety inventory. The digital twin can improve
the utilisation of warehouse by pre-arranging inventory space for up-
coming components. Therefore, the manufacturers can reduce purchase
uncertainties and better cope with urgent incidents by determining the
precise due date for product delivery and its sensitivity based on the
analysis of procurement data, order backlog data and warehouse data.
Dynamic scheduling is another application which copes with un-
precedented conditions. Once customer orders change (e.g. an earlier
delivery date, extra orders, etc.), the manufacturers can rearrange the
procurement network of customised products and reschedule the pro-
duction plan. Price is then recalculated by simulating and quantifying
the impact of changing orders.
Manufacturing stage
The digital twin offers three primary applications in this stage, in-
cluding 1) smart production planning, 2) smart quality execution and 3)
smart cost evaluation. The processing routine of customised products is
more dynamic and sophisticated than that of traditional ones because
the production of customised products often involves the modular de-
sign, non-linear process plans, alternative routings and variable se-
quences [36]. However, the digital twin makes production planning of
customised products more efficient and intelligent. Real-time states of
manufacturing systems (such as resource availability, operating con-
ditions, equipment efficiency, etc.) are obtained and represented in the
virtual entity. With access to tremendous data, big data analytics can
pair all available resources regarding their available time and capa-
cities. Besides, a series of scheduling algorithms are adopted to de-
termine the optimal production processes and sequences, required
manufacturing facilities and the production schedule.
As for smart quality execution, the digital twin can serve the all-
around quality monitoring, early warning of quality defects and rapid
diagnosis of root causes through accessing a variety of quality-related
data (such as tool wear, geometry and machining parameters). For in-
stance, the correlation between the registration accuracy of the printing
machine and the geometry of impression rollers can be modelled and
the defective printing can be predicted via continuously monitoring of
these parameters. Once the geometry of rollers deviates from accep-
table tolerance, an early warning implying preventive maintenance will
be sent to equipment engineers. Besides, historical data will be stored
and used to identify the most influential parameters and their accep-
table ranges. Once crucial quality metrics and their root causes are
defined, the manufacturing system will be able to monitor, diagnose,
and correct operating conditions of machines on time. As a result,
product defects will be controlled within an acceptable range.
Traditionally, costs are calculated in batch size and distributed
evenly to each product. In comparison, the digital twin can calculate
the costs of a product in the unit level. With the help of various sensors
installed in the manufacturing systems (e.g., machines, warehouses, air
conditioning systems),the resource consumption (such as electricity and
gas consumption, and tool wear) of each operation can be calculated.
For instance, the digital twin can calculate the milling cost of each
component that goes through a specific milling machine by metering
this machine’s electricity consumption. As a result, the cost of each
product is determined to help develop a more customised and evidence-
based pricing strategy.
Usage stage
In the usage stage, the digital twin breaks the information barrier
Fig. 4. Digital twin-augmented data-driven smart customisation applications.
X. Wang, et al. Journal of Manufacturing Systems xxx (xxxx) xxx–xxx
7
caused by geographical isolation, enabling service providers to con-
tinuously monitor product status and working environments-such
monitoring guides designers to uncover more customer voices and offer
highly personalised services to increase customer satisfaction. The di-
gital twin module in the usage stage mainly aims at delivering value-
added services to customers by analysing the real-time status of pro-
ducts and working environments. The applications of the context-aware
service and smart service recommendation are involved in this stage.
Context-aware product services supported by big data from the di-
gital twin module are highly expected by customers. The sensor net-
work enables products to capture multi-sourced context information.
Meanwhile, big-data analytics supports the requirement mining pro-
cess. Once customer requirements are identified, the control systems,
software systems and informatics of the products are adjusted so as to
fulfil customer requirements. Besides, the user environment digital twin
can facilitate the intelligence of product when customers are interacting
with components, hardware and software. Based on big-data analytics
and machine learning, products can automatically learn the patterns
between environments and user behaviours and thus adjust themselves
for the convenient use of customers. As a result, the service providers
who are informed of the interaction between products and user en-
vironments can detect emerging customer requirements and provide
value-added services. For instance, a coffee machine can adjust its
heating temperature autonomously by detecting the outdoor tempera-
ture, humidity and weather, thus keeping coffee perfect in taste.
Smart maintenance service recommendation aims at optimising
product performance during operation to make it adjustable to the
using environment. The digital twin module is employed for the fault
diagnosis of mechanical and electrical systems. Various data (such as
material characteristics, operating conditions, and environments) are
obtained to measure the failure mode and degradation state of the
physical system. The digital twin is constructed to simulate and predict
normal and abnormal behaviours of the asset under different circum-
stances. Once the virtual entity is well trained, data analytics is em-
ployed to collect, process and analyse fault characteristics as well as
predict early warnings of physical assets. Under this circumstance, the
digital twin can quickly anticipate the service life of the product and
develop proper preventive maintenance for customers. Data analytics is
kept being trained by troubleshooting records (such as failure modes,
health indicators, diagnostic rules, etc.) stored in the knowledge data-
base. Subsequently, the digital twin can proactively recommends per-
sonalised maintenance services to customers through data analytics.
The implementation of the digital twin effectively prevents human
errors and accidents from occurring in the installation process. Through
collecting a variety of data about environmental factors that affect
human behaviours, the digital twin can identify, predict and prevent
human errors such as incorrect sequencing, improper spatial arrange-
ment and other management problems proactively, thereby reducing or
even eliminating human errors related to on-site quality control.
Case Study: Digital Twin-Augmented Smart Elevator
Customisation
In this section, a case study on the application of the proposed
framework to an elevator manufacturer is conducted. In this case study,
common issues in elevator customisation are analysed and potential
solutions are proposed based on the framework outlined above.
Elevators are highly customised products for specific buildings
considering such factors as the building structure, passenger flow, using
purposes, expected usage periods, etc. Recently, the evolvement of ur-
banisation has made the functionality of buildings more complicated
and diversified. The emerging city-complex (also known as HOPSCA, a
multifunctional building that provides diversified services such as ho-
tels, offices, parks, shopping malls, conventions, and apartments) brings
new opportunities as well as challenges to elevator industries. Different
from traditional elevators which are designed only for single-purpose
buildings, the elevators for modern city-complexes are expected to meet
the requirements of a more sophisticated internal environment as fol-
lows:
• To assist the connection and diffluence of diversified customer
bases: A modern city-complex is usually featured by providing
multi-phase services (e.g., retail stores, entertainments, offices, re-
sidential apartments, winter gardens, and hotels) in a single building
to support various visiting purposes of tens thousands of resident
populations. The increasingly sophisticated customer behaviours
within the building necessitate well-functioned internal transit
networks that allow unrestrained pedestrian movement in the
modern city-complex.
•To be adaptable to more unique and personalised building archi-
tecture: Besides the functionality of buildings, some basic para-
meters of elevators (e.g., structure, length and width of the shaft,
height of the engine room, and height of the last stopping point)
vary significantly among different buildings. Each miscalculation of
dimensions can lead to a 10% additional final cost, and the figure
will become larger when buildings are getting more sophisticated.
However, the present poor integration between elevator manu-
facturers and building companies causes much unnecessary re-
working and project delay.
• To fit in with stressful production planning: Elevator construction is
a budget-intensive project, of which the completion time ought to be
synchronised with the building construction schedule tightly. In
order to better fulfil customer requirements, manufacturers should
adjust their production plan to the real-time condition of the site.
• To avoid onsite installation mistakes and reworking: Installation
mistakes such as lack of integration between the elevator and other
building subsystems (e.g., structure, walls, electrical services, etc.)
and mistaken installation sequences can lead to price recalculation.
The industry has attached great importance to avoiding reworking
caused by mis-operation in the complicated and time-consuming
installation activities.
• To deliver more personalised services: Nowadays, elevators have
become an integral part of the service ecosystem within the
building. New service requirements for elevators have been put
forward recently, such as VIP services, access restrictions, nonstop
travel, etc. It is foreseeable that elevator services are going to be
tailored to each passenger in the future.
To sum up, modern buildings call for elevators smarter in design,
flexible in production and personalised in service, and also require
customer collaboration throughout the whole customisation activities.
The digital twin-augmented smart customisation brings great benefits
to both elevator manufacturers and customers. For manufacturers, in-
creased data availability and real-time simulation can improve re-
quirement fulfilment and resolve potential risks of products. For users,
it can support real-time updates according to the operating conditions.
The real-time data can also provide reference for future research on
designs. Besides, the digital twin of elevators can integrate with smart
building systems or even smart city systems to achieve a more sus-
tainable future city.
As shown in Fig. 5, the elevator customisation system involves three
primary stages, which are elevator designing, collaborative production,
and the personalised service. The physical entities of the system include
manufacturers, the elevator and the building where the elevator oper-
ates. Various data are generated continuously in the three stages. Big
data analytics carries out intelligent material assignment, tracking,
predictive maintenance and energy-efficient management using the
data from the manufacturers (manufacturers’ and suppliers’ production
plans, manufacturing resource utility and machine running state) up-
loaded to the big data centre. Data from the elevator that reflect the
real-time conditions throughout the entire lifecycle of the elevator en-
able the manufacturers to track material distribution, quality and
X. Wang, et al. Journal of Manufacturing Systems xxx (xxxx) xxx–xxx
8
health conditions of the elevator through big data analytics. For in-
stance, the dynamic scheme of material location, batch, type and
quantity can be obtained by capturing and reading information in RFID
tags embedded in required modules of the customised elevator. In the
proposed framework, data from the building environment (e.g. the
number of occupants in the building, weather, and customer identities)
are collected to pre-identify real-time customer demands.
Physical entities and virtual entities are connected and synchronised
to expand customisable objects through a series of simulation and vi-
sualisation technologies. There are three virtual models involved, in-
cluding the elevator model, the manufacturer model and the environ-
ment model. The elevator model reflects operations of the elevator
throughout its lifecycle. The environment model is the high-fidelity
replication of the building environment, which shows how the elevator
interacts with its surrounding environment. Data from the building,
natural environment, interior environment and mobile devices are
collected to reflect dynamic changes of the context. Apart from elevator
and environment models, the manufacturer models are built to re-
present the manufacturing systems, the supply chain, resource condi-
tions and production progress. Different models are integrated to si-
mulate possible interactions in the virtual space. The combination of
the environment and elevator models, for example, can simulate how
the operating elevator affects its surrounding environment. The re-
lationship patterns derived from virtual space of the digital twin will
facilitate customisation decisions in turn. Data analytics, machine
learning algorithms, simulation models and predictive models of the
digital twin collect, integrate, analyse and visualise multi-sourced data
accumulated in the big data centre to optimise customisation activities.
As a result, the elevator is becoming more customised in design, flexible
in manufacturing and personalised in service. Three detailed cases are
illustrated below.
Customised Noise Control
The digital twin facilitates further participation of customers in the
elevator customisation in the virtual space. For instance, elevator noise
control will become a customisable option. Traditionally, elevator noise
is primarily tested under the laboratory environment. However, testing
data cannot be applied to real buildings because actual noise is highly
dependent on the building structure, construction materials and ele-
vator assembly techniques. With the help of the digital twin, customers
and manufacturers can establish an on-site elevator noise model, and
hence optimise elevator designs without having to integrate systems
physically. To establish an elevator noise prediction model, a variety of
noise characteristics and sources (e.g., elevator architecture, frequency
response function, forces, and stiffness) are defined, the airborne noise
data of elevator components (e.g., machine drive, guide rail, pulley,
landing doors, car buffer) are collected through lab testing, and the
structural characteristics of the building are also identified in the
building information model (BIM) by customers. Driven by the digital
twin, customers can predict the vibration pass-way and airborne noises
with the help of a system coupled by elevator noise models, component
models and the BIM. Hence, manufacturers and customers can adjust
design parameters accordingly, which will significantly reduce in-
tegration issues in elevator installing.
Customer-driven Production Plan
Elevator’s production plan is tightly subject to construction progress
(Hoistway is built as the building rises, and the machine should be
equipped before building is topped out). This tight schedule requires a
production plan to be more responsive and have a better forecast cap-
ability. The digital twin enables manufacturers to test the machine
settings in the virtual space before physical changeover through a si-
mulation model based on data generated from humans, machines and
environments in the shopfloor, thus significantly shortening the ma-
chine setup time in producing customised elevators. Besides, the model
will significantly reduce planning time required for customised product
under small batches.
Data generated by the digital twin will also reduce uncertainties
Fig. 5. Digital twin-augmented elevator customisation.
X. Wang, et al. Journal of Manufacturing Systems xxx (xxxx) xxx–xxx
9
caused by machine downtime. Machine breakdown was previously re-
garded as a discrete probability event. However, now manufacturers
can monitor machine conditions in real-time and predict machine
breakdown and bottlenecks in advance with the help of IoT technology
and data analytics. The digital twin dynamically adjusts production
routes and schedule resources to reduce disturbance. Moreover, man-
ufacturers can forecast the precise date when the building is ready for
elevator installation by monitoring the construction process. With the
reduction of uncertainties, manufacturers can dynamically order raw
materials from suppliers, schedule resources and arrange production
activities.
Customised Service for Energy Efficiency
Currently, mainstream service providers only offer standard services
to users, but diversified customer requirements resulted from sig-
nificantly varied operating scenarios of elevators necessitate more
personalised services. With the help of the digital twin, service provi-
ders can model the using-context of elevators, and hence develop cus-
tomised energy-saving strategies. Sensors, cameras and Wi-Fi are
equipped in the building to detect a variety of user data, including the
number of passengers, passenger profiles, and even customer identities,
which enable the building managers to obtain peak-time flow, peak
season flow, the purpose of visit and other passenger patterns for a
period of time. The digital twin functions to simulate traffic constraints
under different elevator operating conditions. By changing the number
of operating elevators, their control logics, accelerations and speed
rates, the model can carry out the optimal operation methods to max-
imise energy efficiency. With the aid of a series of machine learning
techniques, the elevator system can even predict people flow and self-
organise their operations in the future. When there are not many pas-
sengers in the shopping area during weekdays, elevators will auto-
matically reduce the number of operating elevators to save energy.
Conclusion and Future Work
As an emerging technology, the digital twin will significantly ex-
pand customisation capabilities from the following three main aspects.
Firstly, it enables manufacturers to have a deeper understanding of
their customers and thus results in a better product-service ecosystem
through building a virtual model with real-world data sources.
Secondly, with the help of the digital twin, customers can better par-
ticipate the customisation activities in virtual space, which will tre-
mendously expand customisable objects and facilitate open innovation.
Thirdly, the digital twin renders manufacturing systems more flexible
and predictive, thus reducing risks caused by uncertainties.
However, several limitations should be considered as well. First of
all, as little research mentions user environments of the digital twin, the
data obtained from such models are still limited and single-sourced.
Secondly, previous studies mainly discuss the digital twin at the shop-
floor level or unit level, instead of the whole supply chain level or
customisation organisation level. That being said, this paper serves as a
preliminary exploration of the digital twin-augmented smart customi-
sation framework and its potential applications. For future work, some
promising directions can be pursued by interested researchers:
1) To identify the boundaries of data availability between stakeholders
in the customisation organisation. Considering information from
suppliers or customers is confidential, manufacturers should care-
fully determine the system boundary of the digital twin between
multiple participants. Specifically, what types of data can be shared
within the organisation.
2) To consider the communication between digital twins. The inter-
connection between digital twins of products, manufacturing sys-
tems and environments explores new functions which have not been
realised before. Therefore, future research can focus on the structure
of new digital twin frameworks, required computational meth-
odologies and data fusion methods.
Declaration of Competing Interest
The authors declare that they have no known competing financial
interests or personal relationships that could have appeared to influ-
ence the work reported in this paper.
Acknowledgement
This work is partially supported by the Beijing National Science
Foundation (Grant No. JQ19011).
References
[1] Tseng MM, Jiao J, Merchant ME. Design for mass customisation. CIRP annals
1996;45(1):153–6.
[2] Zhang C, Chen D, Tao F, Liu A. Data Driven Smart Customisation. Procedia CIRP
2019;81:564–9.
[3] Tao F, Qi Q, Liu A, Kusiak A. Data-driven smart manufacturing. Journal of
Manufacturing Systems 2018;48:157–69.
[4] Rosen R, Von Wichert G, Lo G, Bettenhausen KD. About the importance of au-
tonomy and digital twins for the future of manufacturing. IFAC-PapersOnLine
2015;48(3):567–72.
[5] Qi Q, Tao F. Digital twin and big data towards smart manufacturing and industry
4.0: 360 degree comparison. Ieee Access 2018;6:3585–93.
[6] Song W, Sakao T. A customisation-oriented framework for design of sustainable
product/service system. Journal of Cleaner Production 2017;140:1672–85.
[7] Pine BJ. Mass customizing products and services. Planning review 1993;21(4):6–55.
[8] Khan MA, Mittal S, West S, Wuest T. Review on upgradability–A product lifetime
extension strategy in the context of product service systems. Journal of cleaner
production 2018;204:1154–68.
[9] Tao F, Cheng J, Qi Q, Zhang M, Zhang H, Sui F. Digital twin-driven product design,
manufacturing and service with big data. The International Journal of Advanced
Manufacturing Technology 2018;94(9-12):3563–76.
[10] Negri E, Fumagalli L, Macchi M. A review of the roles of digital twin in cps-based
production systems. Procedia Manufacturing 2017;11:939–48.
[11] Jiao J, Ma Q, Tseng MM. Towards high value-added products and services: mass
customisation and beyond. Technovation 2003;23(10):809–21.
[12] Tseng MM, Jiao J. Mass Customization. Handbook of Industrial Engineering:
Technology and Operations Management 2001:684–709.
[13] Uskonen J, Tenhiälä A. The price of responsiveness: Cost analysis of change orders
in make-to-order manufacturing. International Journal of Production Economics
2012;135(1):420–9.
[14] Brown S, Bessant J. The manufacturing strategy‐capabilities links in mass custo-
misation and agile manufacturing–an exploratory study. International Journal of
Operations & Production Management 2003.
[15] Gershenson JK, Prasad GJ. Modularity in product design for manufacturability.
International Journal of Agile Manufacturing 1997;1(1):99–110.
[16] Anderson DM. Build-to-Order & Mass Customization: the ultimate supply chain
management and lean manufacturing strategy for low-cost on-demand production
without forecasts or inventory. CIM Press; 2004.
[17] ElMaraghy HA. Flexible and reconfigurable manufacturing systems paradigms.
International journal of flexible manufacturing systems 2005;17(4):261–76.
[18] Chen FF, Adam EE. The impact of flexible manufacturing systems on productivity
and quality. IEEE Transactions on Engineering Management 1991;38(1):33–45.
[19] Xing B, Bright G, Tlale NS, Potgieter J. Reconfigurable manufacturing system for
agile mass customization manufacturing. 2006.
[20] Wuest T, Liu A, Lu SCY, Thoben KD. Application of the stage gate model in pro-
duction supporting quality management. Procedia CiRP 2014;17:32–7.
[21] Wang L, Wang XV. Cloud-Based Cyber-Physical Systems in Manufacturing. Springer
International Publishing; 2018.
[22] Wang L, Wang XV. Big Data Analytics for Scheduling and Machining. Cloud-Based
Cyber-Physical Systems in Manufacturing. Cham: Springer; 2018. p. 347–75.
[23] Duffy AH, Kerr SM. Customised perspectives of past designs from automated group
rationalisations. Artificial intelligence in engineering 1993;8(3):183–99.
[24] Zipkin P. The limits of mass customization. MIT Sloan management review
2001;42(3):81.
[25] Panchal JH, Messer M. Extracting the structure of design information from colla-
borative tagging. Journal of computing and information science in engineering
2011;11(4).
[26] Wang G, Gunasekaran A, Ngai EW, Papadopoulos T. Big data analytics in logistics
and supply chain management: Certain investigations for research and applications.
International Journal of Production Economics 2016;176:98–110.
[27] Glaessgen E, Stargel D. 2012, April The digital twin paradigm for future NASA and
US Air Force vehicles. In 53rd AIAA/ASME/ASCE/AHS/ASC Structures. 2012. p.
1818.
[28] Tao F, Zhang H, Liu A, Nee AY. Digital twin in industry: state-of-the-art. IEEE
Transactions on Industrial Informatics 2018;15(4):2405–15.
X. Wang, et al. Journal of Manufacturing Systems xxx (xxxx) xxx–xxx
10
[29] Tao F, Qi Q, Wang L, Nee AYC. Digital twins and cyber–physical systems toward
smart manufacturing and Industry 4.0: correlation and comparison. Engineering
2019;5(4):653–61.
[30] Tao F, Zhang M, Nee AYC. Digital twin driven smart manufacturing. Academic
Press; 2019.
[31] Qi Q, Tao F, Hu T, Anwer N, Liu A, Wei Y, Nee AYC. Enabling technologies and tools
for digital twin. Journal of Manufacturing Systems 2019.
[32] Prakash C, Barua MK. Integration of AHP-TOPSIS method for prioritizing the so-
lutions of reverse logistics adoption to overcome its barriers under fuzzy environ-
ment. Journal of Manufacturing Systems 2015;37:599–615.
[33] Wang J, Ma Y, Zhang L, Gao RX, Wu D. Deep learning for smart manufacturing:
Methods and applications. Journal of Manufacturing Systems 2018;48:144–56.
[34] Hiassat A, Diabat A, Rahwan I. A genetic algorithm approach for location-in-
ventory-routing problem with perishable products. Journal of manufacturing sys-
tems 2017;42:93–103.
[35] Tao F, Zhang M, Liu Y, Nee AYC. Digital twin driven prognostics and health man-
agement for complex equipment. CIRP Annals 2018;67(1):169–72.
[36] Spring M, Dalrymple JF. Product customisation and manufacturing strategy.
International Journal of Operations & Production Management 2000;20(4):441–67.
X. Wang, et al. Journal of Manufacturing Systems xxx (xxxx) xxx–xxx
11
