Production Planning & Control

The Management of Operations

ISSN: (Print) (Online) Journal homepage: https://www.tandfonline.com/loi/tppc20

Real-time data-driven discrete-event simulation

for garment production lines

Woo-Kyun Jung , Hyungjung Kim , Young-Chul Park , Jae-Won Lee & Eun Suk
Suh

To cite this article: Woo-Kyun Jung , Hyungjung Kim , Young-Chul Park , Jae-Won Lee & Eun
Suk Suh (2020): Real-time data-driven discrete-event simulation for garment production lines,
Production Planning & Control, DOI: 10.1080/09537287.2020.1830194

To link to this article: https://doi.org/10.1080/09537287.2020.1830194

Published online: 05 Oct 2020.

Submit your article to this journal

Article views: 34

View related articles

View Crossmark data

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

https://www.tandfonline.com/action/journalInformation?journalCode=tppc20
PRODUCTION PLANNING & CONTROL
https://doi.org/10.1080/09537287.2020.1830194

Real-time data-driven discrete-event simulation for garment production lines

Woo-Kyun Junga, Hyungjung Kimb, Young-Chul Parkc, Jae-Won Leec and Eun Suk Suhd
a
Department of Mechanical and Aerospace Engineering, Seoul National University, Seoul, Republic of Korea; bInstitute of Advanced
Machines and Design, Seoul National University, Seoul, Republic of Korea; cHojeon Limited, Seoul, Republic of Korea; dGraduate School of
Engineering Practice, Institute of Engineering Research, Seoul National University, Seoul, Republic of Korea

ABSTRACT ARTICLE HISTORY

The garment industry, which is traditionally labour-intensive, is facing significant challenges to reduce Received 22 September 2019
production costs in the face of material, labour, and infrastructure maintenance cost increases. To Accepted 23 September 2020
address this challenge, various aspects of smart factory-related technologies are being adopted by the
KEYWORDS
garment industry. In this study, we introduce a production line simulation approach based on real-
Discrete-event simulation;
time power monitoring data in the form of garment production workers’ individual task times. Power real-time data acquisition;
consumption data for workers’ sewing machines are collected using ICT-based power monitoring devi- power monitoring system;
ces, and the data are transmitted to a cloud server through a wireless network and processed by an garment production; smart
approximation algorithm to extract accurate task time. Power monitoring data are then used as inputs factory technology
to a discrete-event simulation model to measure the overall productivity, individual productivity, and
other key performance indices critical to the garment production line. The simulation results were
18.8% more accurate than the results obtained by the garment production line using a conventional
approach. This study demonstrates the potential of smart factory-related technologies and how they
can be implemented in different types of industries, such as the garment production industry.

1. Introduction entered into a worksheet to measure the line productivity.

This current practice has several shortcomings. The most
With fierce global competition, many manufacturing indus-
important issue is the use of a small number of samples (typ-
tries face growing pressure to increase productivity and
ically 3–5) at the garment production site to determine the
reduce internal costs in their respective industries. Recently,
working time of each worker. For this reason, it does not
the notion of smart factory, which integrates Industry 4.0
capture the task time variation of individual workers.
related technologies such as robot technology, artificial intel- Increasing the number of samples allows for more reliable
ligence, big data, and cyber physical systems, has gained data collection, but since this is related to budgetary issues,
attention and is being implemented in industries to improve most garment companies use only a small number of sam-
overall productivity. ples. This will lead to inaccurate estimations of overall prod-
One of the industries embracing the latest smart factory uctivity and line balance, which need to be resolved using
paradigm is the garment industry. This industry is tradition- repeated trials, causing delays in line stabilisation and an
ally labour intensive, as most of the parts fabrication and increase in overall production costs. Although the accuracy
garment assembly is done through manual labour. Owing to of the data is an important part of the simulation, merely
increasing materials and labour costs, while the average price collecting accurate data in labour-intensive industries such as
for garments remains constant, garment firms around the the garment industry is of limited utility. In particular, it is
world are seeking ways to reduce the overall cost and very difficult to collect real-time data that change frequently
increase overall productivity. To this end, garment produc- in the garment production site and then apply it to simula-
tion firms are starting to explore and adopt smart factory- tions. The issue is not simply about the accuracy of the data.
related technologies and approaches. In terms of how to collect data, installing additional physical
One of the key issues in garment production is line bal- hardware that collects data on inexpensive equipment such
ance. The garment production line, as with production lines as sewing machines is inefficient on a budget.
in other industries, needs to be balanced to achieve a con- In order to address the aforementioned issue, we propose
sistent production rate, while preventing bottlenecks and a discrete-event simulation model-based approach to assess
starvations. Currently, garment production lines are designed real-time garment production line productivity using real-
by subject matter experts who rely on their previous experi- time production line data. An individual worker’s real-time
ences. The task time for each work station is estimated and task data are collected using an ICT-based power monitoring
measured with stopwatch several times and the results are system (PMS) installed in sewing machines and transmitted

CONTACT Eun Suk Suh [email protected] Graduate School of Engineering Practice, Institute of Engineering Research, Seoul National University, Room 406,
Building 38, 1 Gwanak-ro, Gwanak-gu, Seoul 08826, Republic of Korea
ß 2020 Informa UK Limited, trading as Taylor & Francis Group
2 W.-K. JUNG ET AL.

to a cloud using wireless network. Transmitted data are then presented a virtual simulation platform with model, commu-
processed through an approximation algorithm and format- nication, and optimisation techniques for the CPS implemen-
ted to be used as inputs to a discrete-event simulation tation of a long production line in a steel manufacturing
model of the garment production line. Using this approach, plant. Brandstetter and Wehrstedt (2018) proposed a flexible
accurate work station task time-based productivity assess- simulation framework that can combine different simulation
ment can be accomplished in real-time, thus identifying models to provide simulations of cyber physical production
potential bottleneck and starvation points and monitoring systems. Lachenmaier, Lasi, and Kemper (2017) considered
the utilisation rate experienced by individual work stations. the application of simulations to CPS through the case of a
To demonstrate the advantage of this approach, simulation car manufacturer. Go €lzer and Fritzsche (2017) summarised
results obtained by this research were compared with results the data requirements and proposed a decision framework
obtained from the conventional approach for the same gar- based on real-time field data.
ment production line. Various simulations related to garment production have
The main contribution of the research presented is the also been studied. Liong and Rahim (2015) simulated the
introduction of a more accurate real-time productivity assess- waiting time between processes within the overall sewing
ment approach for the garment industry that also reduces process using Arena software, and Bevilacqua et al. (2013)
the possibility for human induced error. Subsequent sections undertook a discrete-event simulation of garment production
are arranged as follows. After the review of the relevant according to the flow time and production capacity during
research literature, the overview of the current garment pro- the sewing process. Unal,€ Tunalı, and Gu €ner (2009) con-
duction line design process is presented. The new simula- ducted a simulation of line balancing considering the states
tion-based approach is presented next, followed by a case of sewing lines, and Zu €lch, Koruca, and Borkircher (2011)
study using a real garment production line. The results of simulated a scenario for a garment company’s production
the simulation, a comparative analysis of simulation results logistics using the simulation tool FEMOS. Greasley (2008)
with the conventional approach is presented, followed by a proposed a discrete-event simulation model for the storage
discussion, conclusion, and recommendations for space of an overseas textile manufacturing plant. Dang and
future work. Pham (2016) analysed a footwear assembly line with a hori-
zontal working structure and uncertain task times through a
discrete-event simulation. Perona et al. (2016) studied how
2. Literature review
to reduce the lead time between sewing operations using
The aim of this study is to introduce a simulation-based Little’s law. Additionally, Lee et al. (2018) measured the rota-
approach that reflects the real-time status of a garment pro- tion of the bobbins of sewing machines using ICT-based sen-
duction line using an ICT-based PMS. This not only allows for sors and presented the plant operation status through a
a more accurate analysis of the output than traditional simu- discrete-event system simulation-based on the meas-
lation methods but also identifies other key performance ured data.
indices, bottleneck regions and starving regions that change On the other hand, there is ongoing research regarding
over time. Related research includes manufacturing process garment production task time. The time to make a garment
simulations, task time measurement methods and a power can be summed up by the value known as the standard
consumption monitoring study. minute value (SMV), which is the task time for each unit pro-
Manufacturing processes have been simulated in various cess. Balancing a production line can be expressed as the
fields. Most manufacturing process simulations primarily SMV balance of each operation. Hanan and Seedahmed
focus on optimising the process schedule or on assembly (2018) studied how to improve line efficiency by predicting
line balancing (Jeon and Kim 2016, Mo €nch and Zimmermann an efficient SMV in the manufacturing process of long-
2011). Nejad, Sugimura, and Iwamura (2011) conducted simu- sleeved shirts, and Howlader et al. (2015) proposed a method
lations to optimise process plans and schedules for auto- to increase the efficiency of a T-shirt production line by mini-
mated machine tool production systems and verified the mising the SMV through the matching of manpower, equip-
results through experiments. ment, materials, and methods. However, an analysis of the
The research into applying real-time measurement sensor previous works mentioned shows that they are limited in
data from the physical world to simulations was actively dis- that they used task times measured by stopwatch, which do
cussed in the 2000s. Representatively, simulations driven by not represent the overall production time owing to the
field data were run through Symbiotic simulation. Aydt et al. aforementioned human error and the paucity of measure-
(2008, 2009) defined symbiotic simulations and categorised ment samples taken.
them according to their problems, and suggested further With the advancement of sensors and ICT-based technolo-
research directions, such as the need for an optimisation pro- gies, various studies on power consumption monitoring
cess. Onggo et al. (2018) proposed the implementation of a using wireless communication have been conducted (Zhou
digital twin by applying the latest technologies such as et al. 2017; Juhana and Irawan 2016; Chen et al. 2011). For
machine learning and big data to a symbiotic simulation garment production line related work, power monitoring
model (Aydt et al. 2008). research mostly concentrates on the efficient use of sewing
In recent years, a variety studies involving cyber physical machine energy for production purposes (Yoon et al. 2020;
systems (CPS) have been conducted. Iannino et al. (2019) Muthukumarana et al. 2018; Çay 2018; Habib et al. 2016;
 PRODUCTION PLANNING & CONTROL 3

Rogale et al. 2005). There are few studies that analyse the
patterns of sewing tasks through power monitoring and
extract time data using the findings. The most important fea-
ture of the simulation in this study is that the data for the
production line were collected real-time using an ICT-based
data source, and extracting results using the data through
the simulation model. Robertson and Perera (2002) emphas-
ised data accuracy and the sources, systems, and timelines
used in simulations, and Wang et al. (2011) undertook simu-
lations based on data collected onsite to distribute materials
efficiently for a rapid analysis and for the line redesign of an
automotive production line.
As seen from the works surveyed, a variety of research on
garment production lines, garment production time, and
real-time production line data acquisition has been pub-
lished. However, there is very little research regarding the
use of real-time power monitoring data to assess the overall
productivity of a garment production line. Lee et al. (2018)
proposed a simulation model that utilises a sewing
machine’s bobbin spinning measurement data acquired
through ICT technology. However, the aforementioned work
was not able to simulate the entire garment production line Figure 1. Production process of a garment production line.
real-time using dynamic data. Compared to the results pub-
lished by the previous literature, the approach proposed in order, including the design, material, and production quan-
this study differs in that it utilised the entire work force’s tity for a specific garment, is received and confirmed, the
task time data in real-time, and through this, we were able production planning department analyses the process based
to monitor an individual worker’s workload, bottleneck on the SMV and the work know-how of subject matter
regions, and other key performance indices critical to gar- experts, subsequently ordering each process and forming the
ment production. production line. Owing to the close causal relationship
between the production line design, production cost, and
3. Simulation model based on real-time data delivery date, designing the final production line properly to
acquisition from the power monitoring system ensure maximum productivity is crucial.
The production department arranges the workers in each
The main purpose of this study is to introduce a simulation- process in the production line based on the skill level of the
based approach to assess garment production line productiv- individual workers. Once the workforce is assigned to a par-
ity using task time data collected in real-time from sewing ticular work station, the position of manufacturing equip-
machines in the production line. To this end, an ICT-based ment, such as the sewing machines, is adjusted before
PMS was developed and deployed to collect the power pro- production starts. Bottleneck and line starvation phenomena
file data generated by the sewing machines on a garment that occur during the production process are eliminated by
production line, and an individual worker’s task time was the adjustment of the workforce or the input of a reserve
extracted by processing the collected power monitoring data work force. When the garment production is complete, it is
through a pattern analysis algorithm. The obtained time data transported to the delivery site and the garment production
were converted into hourly intervals of time and reflected in
line repeats the style change process of relocating the equip-
the simulation to check the status of the garment production
ment for the production of the next garment product.
line in real-time.
This section describes the flow of the garment production
line and the simulation modelling method. It also summa- 3.2. Garment production line modelling
rises the process of collecting and processing field data
through PMS. Since the basic purpose of simulation is to realise the phys-
ical phenomenon of a production site as a digital twin on a
computer, verification of the generated model is an essential
3.1. Current design and production process for garment process. With this in mind, the study modelled and validated
production lines the configuration of a variety of lines, ranging from simple
The garment production system consists of raw material sup- serial lines to assembly lines and closed repeat serial lines. In
plies, subsidiary materials, cutting departments, warehouses, addition, three rather simple production lines were modelled
and several garment production lines that are central to the and verified. In this process, it was recognised that individual
garment production process. Figure 1 shows a typical pro- process time had the greatest influence on the reliability of
duction process of a garment production line. When an the simulation model of the garment production line.
4 W.-K. JUNG ET AL.

form new components. The performance of individual opera-

tors sometimes changes, and these changes affect the out-
put of the entire line. In addition, the productivity,
bottlenecks, worker workload and availability of the entire
garment production line are limited by analysis with a simple
model. Considering the complex nature of the garment pro-
duction line, DES is suitable for simulating garment produc-
tion lines.
The simulation model was constructed using AnyLogicTM
to design the inflow of fabric pieces into the target garment
production line, the sewing work of each worker, and the
movement of the combined module fabrics in the line. After
that, the work time of each worker was built into the data-
base and reflected in the simulation model. The simulation
model in this study attempts to reflect the appearance and
phenomena of the garment production line as closely
as possible.
Mak, Wong, and Leung (2014) suggested the first-in-first-
out (FIFO) concept, an abundance of raw materials, the
multi-tasking capabilities of workers, and no impact of
machine downtime as assumptions for line-based production
process simulations. For the simulation model, several
assumptions regarding the garment production line were
made. These assumptions were discussed with garment pro-
duction line managers and were agreed upon for incorpor-
ation into the simulation model. The work performed by
each worker occurred sequentially for each worker (FIFO).

1. Raw materials for sewing tasks (cut pattern) are always

available so as not to affect the worker’s sewing task.
Figure 2. Garment production process flow for the target production line. (On the actual garment production line, the cutting pro-
cess is always done by establishing a cutting plan so
To build the garment production line simulation model, that the cutting is not short.)
the process flow of the target garment production line was 2. Actual workforce daily schedules for the production line,
analysed and is shown in Figure 2. The target garment pro- including breaks, were incorporated into the simulation.
duction line consists of 37 sewing operators, five inspectors 3. Machine downtime is not taken into account (the ICT-
and assistant personnel who produce duck-down hooded based PMS does not consider downtime separately
jackets. The jackets consist of a sleeve and cuff, a collar, a because the data are based on the operation of each
front and back, and pockets, among other components as sewing machine).
modules and are completed through a joining and assembly
process. The target line studied here utilises a complex pro-
3.3. Real-time data acquisition for task time
cess consisting of a combination of modular and lin-
through PMS
ear processes.
Figure 3 shows the simulation target line in a real factory. For garment production lines, which are labour intensive, it
The rectangular cell means a sewing machine and an individ- is difficult to predict the production quantity owing to the
ual worker (the number in the cell refers to the number of uncertain production situation compared to an equipment-
individual workers), and the arrow figure shows a position based production line. This is owing to differences in skill
where the cut pieces of fabric are put. The black line shows levels and the personal characteristics of workers. In addition,
the flow of the sleeve and other elements in the line, the if bottlenecks occur on the line, it is difficult to check for
grey line shows the backward, and the dotted line shows the production delays before work is completed. Therefore, the
flow of the forward component. For detailed production line designing a garment production line that can increase pro-
modelling, commercial simulation software (AnyLogicTM duction efficiency without increasing the workforce is of
Version 8) was used. The garment production line is mod- great interest to garment production managers. The current
elled as a discrete-event simulation (DES) model since each standard procedure for determining the task times of a sew-
unit task is performed independently and a series of proc- ing worker is to measure with stopwatches the time to pro-
esses occurred continuously (Borshchev 2013). The garment cess, or the time taken for each work station, to calculate
production line is not only introduced with more than 100 the SMV and then to determine the average time for each
fabric patterns in a complicated order but is also merged to station. The time data thus calculated are applied to the
 PRODUCTION PLANNING & CONTROL 5

Figure 3. Schematic diagram of the target production line.

standard task time of each worker. However, this method production at 92% level. In this study, baseline simulations
can include human error, as it is a person taking time meas- used this power data. The simulation results using the
urements for work stations in the production line. Another dynamic time data of each worker’s sewing work extracted
drawback is the limitation of only a partial timeframe for the through the pattern recognition algorithm were compared
time measurement, which cannot properly reflect the charac- with the simulation results using the existing time data
teristic information of workers, who tend to change from measured by humans using a stopwatch.
time to time. More importantly, the number of SMV calcula-
tion tasks is only a static phenomenon in the overall gar-
ment production process because only a small sample is 4. Simulation using real-time data processed by an
applied to the total number of tasks. Therefore, it is impos- ICT-based PMS
sible to reflect the various situations occurring on a produc- The PMS described in Section 3.3 was installed on the gar-
tion site in real-time in this manner. ment production line selected for the simulation, and the
In order to overcome this problem, PMS is deployed in task times for all 37 line workers were extracted by analysing
the garment production line to collect the power data of the collected data. The simulation was performed for the
sewing machines and to analyse the collected data to extract entire production period of the target garment for about
the real-time task time data for each worker. Figure 4 shows three weeks. In this study, the simulation results were ana-
the PMS concept implemented for this study. A power moni- lysed by selecting a specific date after the production line
toring device with a current sensor and a Wi-Fi communica- was stabilised. A simulation of the sewing line was carried
tion module is installed onto each sewing machine to out for each of the following three cases and the results
monitor power consumption of the sewing machine during were compared with the actual production results.
the sewing task. The power consumption data collected are
then transmitted to a cloud server over a wireless network, Case #1: Apply the time measured using the PMS as a time unit,
and the data stored on the server are processed for extrac-
Case #2: Apply the average time measured using the PMS, and
tion of each unit task time and quantity through an algo-
rithm that recognises work patterns from a power Case #3: Apply the time measured using a stopwatch
consumption profile.
The task time was calculated using an approximate statis- An example of time measurement for Cases #1–#3 are
tical algorithm and the similarity measurement method. After shown in in Figure 5. Case #1 represents the real task time
segmenting the designated section through piecewise aggre- of a single worker throughout the day. Case #2 shows a sin-
gate approximation (PAA) and simplifying it to the average gle value representing the daily average task time of a single
value of the divided section, each segment value is repre- worker shown in Case #1. Case #3 represents a single task
sented by a character through symbolic aggregate approxi- time for a single worker in Case #1, which is measured using
mation (SAX) (Lin et al. 2003). The task is then recognised by a stopwatch. The garments to be simulated were produced
measuring the similarity between the work pattern reference for a total of 22 working days over a period of about one
profile represented by SAX and the SAX result of the data month. The simulation ran on day 16 after the start of pro-
collected in real-time. A similarity measurement was applied duction. This was done since that specific working day was
to dynamic time warping (DTW) (Eamonn and Chotirat 2005). judged as the time when the production line was
The method of applying real-time data, as described in this fully stabilised.
study, is similar to the method proposed by Jung et al. Figure 6 shows the simulation progress of a garment pro-
(2019). The results of the power data of sewing machines duction line model. The upper part of Figure 6 shows the
analysed using SAX and DTW were in agreement with actual overall line as the simulation is running, and the lower left
6 W.-K. JUNG ET AL.

Figure 4. Conceptual diagram of the power monitoring system for sewing task time measurement.

Figure 5. Task time configuration by simulation cases.

dashed line area shows enlarged operators 16 (SEW_16 in displayed on each worker’s resources in Figure 6 represents
figure) and 17 (SEW_17 in figure). Operator 16 receives the the workload each worker receives. In the enlarged figure,
cut fabric material, sews, and afterward sends it to Operator both operators 16 and 17 are shown in red with 99% of the
17, who adds new cut material to the workpiece received workload. Meanwhile, in the lower right corner of Figure 6, a
from Operator 16 sews it, and sends it to Operator 18 after- diagram of the cumulative garment production quantity as a
ward. In this model, the operator working with the cut fabric function of time is shown.
material reflected the queueing function before the work It was possible to measure the workload for each individ-
module. The colour code (green, yellow, orange, and red) ual line worker for each hour. The analysis of the workload
 PRODUCTION PLANNING & CONTROL 7

Figure 6. Simulation of the garment production line.

allows an assessment of whether the work distribution of Case #3 on the right side of the figure shows the time
each worker on the line is efficient. The results for each case measured by the stopwatch according to the conventional
applied to the simulation will be explained in detail in a sub- method. It can be seen that the unit task time of each
sequent section. worker is concentrated in the range of 90–140 s. The narrow
spacing of each point can be seen as a good indication of
line balancing for the efficiency of the line. In Case #3, owing
5. Results and discussion to the limited number of sample measurements, the usability
In this section, the task time extraction results of the target of these results as dynamic data that can reflect the time
line workers using PMS are compared with the task times changes is limited. Comparing Case #2 and Case #3, it can
measured by the conventional method, and the simulation be seen that the unit task time gap for each worker is more
results for three different time input cases mentioned in pre- widely distributed in Case #2 than in Case #3. As we
vious section are presented. Additionally, the results of the explained above, line balancing works well when the unit
workload analysis for individual workers are summarised, fol- task time gap of each worker is narrow (Case #3). However,
lowed by discussion on research results and how the frame- the data of Case #2, which measured the task times of real
work and model developed in this study can further improve workers, were quite different from the data of Case #3. These
decision makers’ abilities to manage their produc- results indicate that Case #3 with the stopwatch measure-
tion planning. ments contained human errors or that the actual sample
times could not be reflected because of the low sample rate.
From these results, it is deemed that PMS can replace meas-
5.1. Unit task time analysis of individual operators via urement of work time by manpower; however, additional
the PMS consideration is required for conversion to a probability
Figure 7 shows the time data used for the three cases distribution.
described in the previous section. Each of these points repre-
sents the amount of time taken by each worker assigned to 5.2. Analysis of the simulation results
their task. Case #1 shown on the left of Figure 7 is the result
of extracting the unit task time of individual workers in units Each of the three cases was simulated and the outputs from
of one hour through the PMS, and Case #2 shows the aver- the actual line and the simulation results were compared
age value for each worker extracted from Case #1. In Case (Figure 8(a)). The actual number of garments produced was
#1, the line connecting the task times of individual workers 250 on the target date, and the newly introduced simulation
shows the individual characteristics of each worker. Workers model using the real-time data acquired from PMS predicted
without line fluctuations from the start to the end of their 235, showing 94% reliability. The Case #1 simulation results,
work can be considered as consistent performers, but large reflecting the PMS results, and the Case #2 simulation results,
fluctuating curves over time indicate that the workers’ per- reflecting the average time of the PMS results, showed errors
formances were not consistent. Additionally, the graph in of 6% and 12% relative to the actual output, respectively.
Case #1 shows that a number of workers are somewhat less The Case #3 simulation results, which used time data meas-
efficient until 9 AM, stabilising afterward. In Case #1, as ured with stopwatches, showed an error of 24.8%. The
opposed to Case #3, the task time of each worker according results shown in Figure 8(a) effectively demonstrate the feasi-
to the overall time can be used as dynamic data. bility of the newly proposed simulation approach.
8 W.-K. JUNG ET AL.

Figure 7. Comparison of individual worker’s unit task times between the real-time PMS data (Case #1), average time data (Case #2), and stopwatch measured data
(Case #3). Each dot represents average task time of individual operators per hours

Figure 8. Simulation results of the target production line. (a) Comparison of daily production results between the actual production and the simulation results. (b)
Comparison of the time-dependent simulation results between dynamic data (Case #1) and static (Case #2 and Case #3)
 PRODUCTION PLANNING & CONTROL 9

Table 1. Statistical figures and Monte Carlo results of PMS-based simulation results.
Statistical figures Monte Carlo simulation results
Mean 23.50 Conditions 10,000 random numbers
Standard deviation 4.72 Normal distribution
Standard error 1.49 Results Production per hour range: 23.47 ± 9.23 (95% reliability)
Median 24.00

Figure 8(b) shows the simulation results of Case #1 and productivity, as identified in Figure 8(b). On the other hand,
#2, which applied the PMS data proposed in this study over in Case #3 (Figure 9(b)), the six production line workers
time and Case #3, which applied the conventional data col- showed nearly identical workloads, with little change in the
lection method. When the PMS data are applied to the simu- workload over time. Additionally, this method resulted in a
lation (Case #1), it is also possible to check when the low workload rate of less than 50%, which does not reflect
production line has the highest productivity. The line had the reality of the situation as shown in Case #1. Finally, the
the highest hourly production between 1 and 3 pm, while point at which the workers’ workload rates are high can be
the lowest productivity was between 9 and 10 am and 4 and considered as a bottleneck. Bottlenecks are situations where
5 pm. These results cannot be extracted from the results of the workstations that make up a production line continue to
Case #2 or Case #3, in which the average data value or con- receive a load close to the maximum capacity they can
ventional time measurement method is applied. accommodate, thus affecting overall yield. In Figure 9(a),
Based on the simulation results presented in Figure 8, an OP37 receives the most workloads throughout the day. In
additional statistical approach was conducted. In the simula- the 8 am to 10 am section, OP37 is not a bottleneck because
tion using the existing static data, it was impossible to pre- it is loaded within capacity, but for the rest of the time, the
dict the production per unit time, but this stochastic data indicate that OP37 is under the maximum workload.
approach is possible through simulation results based on This situation can be explained by examining operator influ-
PMS data (Case #1 in Figure 8(b)). ences upon production along the line. A bottleneck will
Table 1 shows the statistical numerical data for Case #1 appear in relation to workstation workloads before and after
and the Monte Carlo simulation results. The Monte Carlo a particular workstation. In the current situation, the OP37
simulation is a representative method of the stochastic status only affects OP36. OP35 does not seem to be heavily
approach through random numbers. The simulation using affected by OP36, so it is not a bottleneck, and neither do
PMS data showed a deviation of about 9.23 productions the rest. By eliminating this bottleneck, the productivity of
based on the average of 23.50 productions per hour in the the production line can be improved. Through this study, we
95% confidence range. This means that the line can produce found a bottleneck on the production line through the simu-
up to 32.70 products per unit time based on current produc- lation using the PMS data that was undetectable with the
tion trends. It is necessary to adjust for various variable fac- simulation using the conventional data collection method.
tors, but through this, the maximum production capacity of
the production line can be predicted.
5.4. Discussion
Analysis results revealed several important aspects of the
5.3. Workload analysis of production line workers
PMS-based production line simulation. Additionally, results
When workers engage in a series of tasks required for their show how the proposed simulation framework can be used
workstations, it affects the performance of the workers by managers to improve the overall production line oper-
located ahead or behind. The accumulation of these actions ation through implementation of smart factory technologies
affects the performance of the entire production line. To pre- such as ICT-based PMS and a discrete-event simula-
vent this from happening and to improve the efficiency of tion model.
the line as a whole, production managers need to balance First, compared to traditional stopwatch-based time meas-
the workloads for individual workers working in the produc- urements, the PMS-based measurements of individual work-
tion line by performing a line balancing analysis. ers’ unit task time shown in Figure 7 clearly demonstrate a
Figure 9 shows the workload rates for six production line more accurate picture of each worker’s performance on an
workers working on six successive workstations of the pro- hour-by-hour basis with proper monitoring infrastructure in
duction line modelled. Figure 9(a) shows the workloads of place. It is true that initial investment for the PMS is
workers by time zone in Case #1, which simulates data col- required. However, once the system is in place and operat-
lected by applying PMS data, and Figure 9(b) shows the ing, the actual workers’ unit task times can be observed in
workload rate of Case #3, which simulates data collected by real time, which is not possible with the current measure-
the conventional method using a stopwatch. In Case #1 ment practice. This allows managers to identify where to
(Figure 9(a)), OP 37, the final worker, shows a consistently look for the cause of task time fluctuation and improve it in
high load rate, while OP 33 shows a relatively low load rate, terms of unit task time variation.
although it varies with time. Additionally, it can be seen that Second, discrete-event simulation model of the garment
workers exhibited low overall load rates between 9 am and production line, combined with ICT-based PMS, yielded pro-
10 am and 4 pm to 5 pm, which showed the lowest line duction results that mimic reality more accurately than do
10 W.-K. JUNG ET AL.

Figure 9. Comparison of assembly line operators’ workloads between dynamic (Case #1) and static data (Case #3). (a) Assembly line operators’ workloads for the
simulation result of Case #1. (b) Assembly line operators’ workloads for the simulation result of Case #3.

the results obtained by point data for unit task time, as This type of simulation could be further conducted to opti-
shown in Figure 8. This allows managers to predict hour-by- mise the garment production line work through implement-
hour production quantities more accurately and take neces- ing an optimisation algorithm, yielding optimum workforce
sary measures to control stable production rates of the placement on the garment production line that can achieve
garment production line. maximum productivity with proper line balancing and bottle-
Third, results shown in Figure 9 provide important infor- neck control.
mation for the production line operation. This figure identi-
fies workers with light and heavy workloads throughout the
production shift, allowing managers to identify potential bot- 6. Conclusion and future work
tlenecks within the production line. Thus, managers can
In this study, a discrete-event simulation model-based
either adjust the amount of tasks performed at workstations
approach using an ICT-based PMS was developed to meas-
with a heavy workload or operate target workstations with
ure and analyse task times in real-time for a garment produc-
more skilled and experienced workers, thus reducing the
tion line. The proposed approach was implemented upon an
load and matching the level of worker skill with task diffi-
actual garment production line to collect and analyse the
culty of the station.
The PMS-based garment production simulation model data. Although an appropriate statistical test analysis was
introduced in this study can be a foundation for several not performed, it was verified that the discrete-event simula-
important improvements on how managers do their work. tion results of the garment production line using secured
Using PMS, work patterns for individual workers can be deterministic data were 94% similar to the actual production
accurately identified, enabling a more effective management results. This is 18.8% higher than the results obtained by the
of the workforce. Additionally, although it was not done in conventional method. Additionally, the proposed approach
this study, the level of work difficulty for each workstation was able to extract a variety of information that was impos-
could be further identified, and in parallel, the skill level of sible to obtain when conventional methods are used. The
each worker could also be identified. This could be further simulation model-based approach, utilising real-time data
implemented in the simulation model in order to place work- obtained by PMS, will be able to replace the manual task
ers with the correct skill level at each individual workstation. time measurement method, and it will enable analyses of
 PRODUCTION PLANNING & CONTROL 11

situations and production efficiency rates more accurately on CEO. Young-Chul Park is the Chairman of Hojeon
a production line. Future work to follow up on this study, as limited Corp., which has 5 global corporations and
mentioned in the case study discussion, includes devising a over 17,000 people working in Indonesia. Since its
establishment in 1985, Hojeon has been keeping
line-balancing optimisation method to resolve the identified partnership with global outdoor/sports fashion
bottlenecks and analyse the big data collected by the PMS. brands such as The North Face, Under Armour,
Additionally, it will be possible to develop a model that can Lululemon, etc. He won Premier Award at model
predict the performance of each worker by analysing the Entrepreneur in 2009 and has various national
award-winning careers as CEO. Hojeon will become
work characteristics of individual workers according to the
an innovative fashion design company by transition from an OEM-ODM
time, day of the week, and specific tasks. specialised company.
However, data collected by the PMS and applied in this
study are deterministic data, and the part that cannot be
applied by converting it into a stochastic distribution is a Dir. Jae-Won Lee is a managing director of Hojeon
limitation of this work, which can be supplemented through limited company, which is one of large apparel man-
future studies. Moreover, additional research is required as ufacturers. He is in charge of Research and develop-
ment for the garment smart factory solution working
the comparison between the output result of the simulation
with the Seoul Nation University from 2017. Before
model and the actual production volume was not performed he join at Hojeon, he had been working at KT for 17
through a statistical method. The stochastic approach can years not only as a business director of new business
robustly guarantee the simulation results proposed in model development team but also a technical engin-
eer in R&D centre of KT. He received a master
this study.
degree in Electrical and Computer Engineering from the Seoul National
University in 2004.

Disclosure statement Dr. Eun Suk Suh received the Ph.D. degree in engin-
eering systems from Massachusetts Institute of
No potential conflict of interest was reported by the author(s). Technology, Cambridge, U.S.A., in 2005. He is cur-
rently an Associate Professor at Seoul National
University. His research interests include system
Funding architecture design, system complexity assessments,
product design, and technology infusion analysis. Dr.
This work was supported by the SNU-Hojeon Garment Smart Factory
Suh’s previous professional experiences includes
Research Centre funded by the Hojeon Ltd. under Grant [number chassis design engineering at Hyundai Motor
0423-20170066]. Company and system architecture research and development at Xerox
Corporation. He has several international patents and peer reviewed
journal publications in the area of printing system design, technology
Notes on contributors infusion and product platform development.

Dr. Woo-Kyun Jung received a Ph.D. in Mechanical ORCID

Engineering from the Seoul National University in Eun Suk Suh http://orcid.org/0000-0002-4693-9596
2020. He was an army officer of R.O.K. from 1997 to
2017 and retired in 2017 as a lieutenant colonel.
Prior to retirement to lieutenant colonel, he served References
as technical staff, military logistics school instructor,
platoon leader, company commander, and battalion Borshchev, A. 2013. The Big Book of Simulation Modeling: Multimethod
commander. Jung’s research interests are in smart Modeling with AnyLogic 6. Hampton, NJ: AnyLogic North America,
factory, soft robotics, smart/composite materials, and Chap. 3. and 8.
appropriate technology. He published over 20 journal articles in Aydt, H., S. J. Turner, W. Cai, and M. Y. H. Low. 2008. “Symbiotic
these areas. Simulation Systems: An Extended Definition Motivated by Symbiosis
in Biology.” 22nd Workshop on Principles of Advanced and
Distributed Simulation, 109–116. IEEE Computer Society.
Aydt, H., S. J. Turner, W. Cai, and M. Y. H. Low. 2009. "Research Issues in
Dr. Hyungjung Kim received his B.S.E. in automotive Symbiotic Simulation." Proceedings of the 2009 Winter Simulation
engineering from Kookmin University, Seoul, Korea, Conference, IEEE, 1213–1222.
in 2004 and Ph.D. in mechanical engineering from Bevilacqua, M., F. E. Ciarapica, A. Crosta, G. Mazzuto, and C. Paciarotti.
Seoul National University, Seoul, Korea, in 2011. He 2013. “Designing an Efficient Production System: A Case Study of a
is currently working as a principal engineer at Clothing Company.” International Journal of Engineering Business
Samsung Electronics Co., Ltd. Before joining Management 5: 36. doi:10.5772/56842.
Samsung Electronics in 2020, he had worked as a Brandstetter, V., and C. Wehrstedt. 2018. “A Framework for
senior research engineer at Doosan Infracore and Multidisciplinary Simulation of Cyber-Physical Production Systems.”
Doosan Robotics, and a postdoctoral researcher at IFAC-Papersonline 51 (11): 809–814. doi:10.1016/j.ifacol.2018.08.418.
SNU. His research interests include appropriate smart manufacturing, Çay, A. 2018. “Energy Consumption and Energy Saving Potential in
smart factory, machine learning, collaborative robots, machine tools, and Clothing Industry.” Energy 159: 74–85. doi:10.1016/j.energy.2018.06.
3D printers. 128.
12 W.-K. JUNG ET AL.

Chen, F., J. Dai, B. Wang, S. Sahu, M. Naphade, and C.-T. Lu. 2011. Liong, C.-Y., and N. A. A. Rahim. 2015. “A Simulation Study on Garment
"Activity Analysis Based on Low Sample Rate Smart Meters." Manufacturing Process.” AIP Conference Proceedings 1643: 460.
Proceedings of the 17th ACM SIGKDD International Conference on doi.org/ doi:10.1063/14907481.
Knowledge Discovery and Data Mining, 21 August 2011, 240–248. Mak, L. C., W. K. Wong, and Y. S. Leung. 2014. “A Simulation Analysis of
doi:10.1145/2020408.2020450. the Impact of Production Lot Size and Its Interaction with Operator
Dang, Q.-V., and K. Pham. 2016. “Design of a Footwear Assembly Competence on Manufacturing System Performance.” Simulation
Line Using Simulation-Based ALNS.” Procedia CIRP 40: 596–601. doi:10. Modelling Practice and Theory 49: 203–214. doi:10.1016/j.simpat.2014.
1016/j.procir.2016.01.140. 09.008.
Eamonn, K., and A. R. Chotirat. 2005. “Exact Indexing of Dynamic Time Mo€nch, L., and J. Zimmermann. 2011. “A Computational Study of a
Warping.” Knowledge and Information Systems 7: 358–386. Shifting Bottleneck Heuristic for Multi-Product Complex Job Shops.”
Go€lzer, P., and A. Fritzsche. 2017. “Data-Driven Operations Management: Production Planning & Control 22 (1): 25–40. doi:10.1080/09537287.
Organizational Implications of the Digital Transformation in Industrial 2010.490015.
Muthukumarana, T. T., H. P. Karunathilake, H. K. G. Punchihewa, M. M. I.
Practice.” Production Planning & Control 28 (16): 1332–1343. doi:10.
D. Manthilake, and K. N. Hewage. 2018. “Life Cycle Environmental
1080/09537287.2017.1375148.
Impacts of the Apparel Industry in Sri Lanka: Analysis of the Energy
Greasley, A. 2008. “Using Simulation for Facility Design: A Case Study.”
Sources.” Journal of Cleaner Production 172: 1346–1357. doi:10.1016/j.
Simulation Modelling Practice and Theory 16 (6): 670–677. doi:10.1016/
jclepro.2017.10.261.
j.simpat.2008.04.009.
Nejad, H. T. N., N. Sugimura, and K. Iwamura. 2011. “Agent Based
Habib, M. A., M. Hasanuzzaman, M. Hosenuzzaman, A. Salman, and M. R. Dynamic Integrated Process Planning and Scheduling in Flexible
Mehadi. 2016. “Energy Consumption, Energy Saving and Emission Manufacturing Systems.” International Journal of Production Research
Reduction of a Garment Industrial Building in Bangladesh.” Energy 49 (5): 1373–1389. doi:10.1080/00207543.2010.518741.
112: 91–100. doi:10.1016/j.energy.2016.06.062. Onggo, B. S., N. Mustafee, A. A. Juan, O. Molloy, and A. Smart. 2018.
Hanan, O. A., and A. I. Seedahmed. 2018. “Effective Way to Estimate the "Symbiotic Simulation System: Hybrid Systems Model Meets Big Data
Standard Minute Value (SMV) of a U3 Shirt by Using Time Study Analytics." Proceedings of the 2018 Winter Simulation Conference,
Technique.” International Journal of Engineering Sciences & Research IEEE, 1358–1369.
Technology 7 (7): 179–184. Perona, M., N. Saccani, S. Bonetti, and A. Bacchetti. 2016. “Manufacturing
Howlader, M. R., M. M. Islam, M. T H. Sajib, and R. K. Prasad. 2015. Lead time shortening and Stabilization by Means of Workload
“Practically Observation of Standard Minute Value of T-Shirt.” Control: An Action Research and a New Method.” Production Planning
International Journal of Engineering and Computer Science 4 (3): & Control 27 (7–8): 660–670. doi:10.1080/09537287.2016.1166283..
10685–10689. Robertson, N., and T. Perera. 2002. “Automated Data Collection for
Iannino, V., V. Colla, J. Denker, and G. Marc. 2019. “A CPS-Based Simulation.” Simulation Practice and Theory 9 (6–8): 349–364. doi:10.
Simulation Platform for Long Production Factories.” Metals 9 (10): 1016/S0928-4869(01)00055-6.
1025. doi:10.3390/met9101025.. Rogale, D., I. Petruni, Z. Drag Evi, and S. A. F. Rogale. 2005. “Equipment
Jeon, S. M., and G. Kim. 2016. “A Survey of Simulation Modeling and Methods Used to Investigate Energy Processing Parameters
Techniques in Production Planning and Control (PPC).” Production of Sewing Technology Operations.” International Journal of Clothing
Planning & Control 27 (5): 360–377. doi:10.1080/09537287.2015. Science and Technology 17 (3/4): 179–187. doi:10.1108/
1128010.. 09556220510590876.
€
Unal, C., S. Tunalı, and M. G€ uner. 2009. “Evaluation of Alternative Line
Juhana, T., and A. I. Irawan. 2016. "Smart Non Intrusive Power
Consumption Monitoring System." Proceeding of 10th Int. Conf. Configurations in Apparel Industry Using Simulation.” Textile Research
Telecommun. Syst. Serv. Appl. TSSA 2016 Spec. Issue Radar Journal 79 (10): 908–916. doi:10.1177/0040517508093416.
Wang, J., Q. Chang, G. Xiao, N. Wang, and S. Li. 2011. “Data Driven
Technology, 1992.
Production Modeling and Simulation of Complex Automobile General
Jung, W. K., H. Kim, Y. C. Park, J. W. Lee, and S. H. Ahn. 2019. “Smart
Assembly Plant.” Computers in Industry 62 (7): 765–775. doi:10.1016/j.
Sewing Work Measurement System Using IoT-Based Power
compind.2011.05.004.
Monitoring Device and Approximation Algorithm.” International
Yoon, H.-S., J.-Y. Lee, M.-S. Kim, E. Kim, Y.-J. Shin, S.-Y. Kim, S. Min, and
Journal of Production Research: 1–15. doi:10.1080/00207543.2019.
S.-H. Ahn. 2020. “Power Consumption Assessment of Machine Tool
1671629. Feed Drive Units.” International Journal of Precision Engineering and
Lachenmaier, J. F., H. Lasi, and H.-G. Kemper. 2017. “Simulation of Manufacturing-Green Technology 7 (2): 455–464. doi:10.100/s40684-
Production Process Involving Cyber-Physical Systems.” Procedia CIRP 019-00063-y.
62: 577–582. doi:10.1016/j.procir.2016.06.074. Zhou, Z., B. Yao, W. Xu, and L. Wang. 2017. “Condition Monitoring
Lee, J. Y., D.-H. Lee, J. S. Song, J.-H. Park, S.-H. Ryu, and J.-H. Park. 2018. towards Energy-Efficient Manufacturing: A Review.” The International
“Study on Stream Cooperation Connected Smart Manufacturing in Journal of Advanced Manufacturing Technology 91 (9–12): 3395–3415.
Textile Timely Production System.” MATEC Web of Conferences 237: doi:10.1007/s00170-017-0014-x.
01005. doi.org/ doi:10.1051/matecconf/201823701005.. ulch, G., H. I. Koruca, and M. Borkircher. 2011. “Simulation-Supported
Z€
Lin, J., E. Keogh, S. Lonardi, and B. Chiu. 2003. "A Symbolic Change Process for Product customization – A Case Study in a
Representation of Time SERIES, with Implications for Streaming Garment Company.” Computers in Industry 62 (6): 568–577. doi:10.
Algorithms." DMKD’ 03, San Diego, CA, USA, 2–11. 1016/j.compind.2011.04.006.