Proposing a Holistic Framework for the Assessment and Management of

Manufacturing Complexity through Data-centric and Human-centric

Approaches

Dominik Kohr

, Mussawar Ahmad

, Bugra Alkan

, Malarvizhi Kaniappan Chinnathai

Lukas Budde

, Daniel Vera

, Thomas Friedli

and Robert Harrison

University of St. Gallen, Institute of Technology Management, Dufourstrasse 40a, 9000 St. Gallen, Switzerland

Automation Systems Group, WMG, University of Warwick, CV4 7AL, Coventry, West Midlands, U.K.

M.Kaniappan-Chinnathai.1@warwick.ac.uk, lukas.budde@unisg.ch, D.A.Vera@warwick.ac.uk, thomas.friedli@unisg.ch,

Robert.Harrison@warwick.ac.uk

Keywords:

Complexity Management, Manufacturing Complexity, Production, Complexity Drivers.

Abstract:

A multiplicity of factors including technological innovations, dynamic operating environments, and globalisa-

tion are all believed to contribute towards the ever-increasing complexity of manufacturing systems. Although

complexity is necessary to meet functional needs, it is important to assess and monitor it to reduce life-cycle

costs by simplifying designs and minimising failure modes. This research paper identiﬁes and describes two

key industrially relevant methods for assessing complexity, namely a data-centric approach using the informa-

tion theoretic method and a human-centric approach based on surveys and questionnaires. The paper goes on

to describe the beneﬁts and shortcomings of each and contributes to the body of knowledge by proposing a

holistic framework that combines both assessment methods.

1 INTRODUCTION

Complexity is induced through a multiplicity of fac-

tors such as technological innovations, dynamic ope-

rating environments, and globalisation, and is belie-

ved to be one of the main causes of the current chal-

lenges in manufacturing, which are lengthy and cos-

tly system design processes, high life cycle costs,

and the existence of numerous failure modes (Alkan

et al., 2016a). According to (Schuh and Schwenk,

2001), complexity in the context of manufacturing,

can be grouped into two categories: i) internal and

ii) external. Internal complexity mainly occurs as

a result of high product variety due to the need to

meet market demands (Chinnathai et al., 2017; Al-

kan et al., 2016b), whereas external complexity re-

sults from market dynamics, political and institutional

complexities (G

otzfried, 2013). There are three main

dimensions of internal complexity: i) structural (sta-

tic) complexity, ii) operational (dynamic) complexity

and iii) organisational complexity (Lindemann et al.,

2008). Structural complexity is related to the no-

tion of the architecture of the manufacturing system,

which is a network that is composed of a set of inte-

racting components/parts. Operational complexity is

driven by the manufacturing systems operational cha-

racteristics (Alkan et al., 2016a). Accordingly, a ma-

nufacturing system can be deemed complex, if its be-

haviours are difﬁcult to describe or predict effectively

(Calinescu et al., 1998). It should be noted that, sy-

stem behaviours are often connected to the underlying

system architecture, hence dynamic complexity has a

strong positive correlation with structural manufactu-

ring system complexity (Sinha et al., 2017). Organi-

sational complexity, on the other hand, is manifested

in organisational structures, systems, processes and in

communication ﬂows (Kohr et al., 2017).

An increase in complexity results in various pro-

blems including: production bottlenecks, reliability

issues and a lack of stability (Efthymiou et al., 2016).

Therefore the management of complexity is a non-

negligible aspect of an organisation’s operation and

thus, a strategy should be in place in order to remain

competitive (Budde et al., 2015). One of the key ele-

ments in effective management of complexity is its

assessment, which should provide a clear picture of

the underlying problems (Alkan et al., 2017). Howe-

ver, the diverse causes and effects of complexity are

hard to evaluate. Furthermore, the methods chosen to

manage them should be chosen based on the indivi-

Kohr, D., Ahmad, M., Alkan, B., Kaniappan Chinnathai, M., Budde, L., Vera, D., Friedli, T. and Harrison, R.

Proposing a Holistic Framework for the Assessment and Management of Manufacturing Complexity through Data-centric and Human-centric Approaches.

DOI: 10.5220/0006692000860093

In Proceedings of the 3rd International Conference on Complexity, Future Information Systems and Risk (COMPLEXIS 2018), pages 86-93

ISBN: 978-989-758-297-4

dual complexity drivers of the company.

In this research, two complementary methods for

assessing manufacturing complexity have been des-

cribed in detail. The ﬁrst method is an information

theoretic measure developed by (Frizelle and Wood-

cock, 1995) deﬁning complexity as the uncertainty

in identifying the required information to deﬁne the

overall state of a manufacturing system. The second

method is a survey based approach identifying manu-

facturing/organisation complexity based on elicitation

and aggregation/pooling of expert opinions. The ad-

vantages and disadvantages of the selected methods

have been identiﬁed based on a number of criteria,

and the authors provide insight for future researchers

of this topic. The combined use of a human-centric

and a data-centric approach is further discussed and a

framework exploiting the advantages of both methods

is proposed.

2 INFORMATION THEORETIC

APPROACH

A practical approach to quantify complexity based on

an entropic model of a factory was developed by (Fri-

zelle and Woodcock, 1995). In this method, the follo-

wing assumptions are made: ”each sub-system is an

immigration-emigration process, reliability is inver-

sely proportional to complexity, cycle time is directly

dependant on complexity and complex processes have

more probability to become bottlenecks” (Calinescu

et al., 1998). Accordingly, two types of complexity

are identiﬁed, i.e. static and dynamic.

2.1 Static Complexity Assessment

Static complexity emerges as a result of the impact

of the product on the resource domain. It is deﬁned

as the amount of information required to deﬁne the

state of the production system (S) and is formulated

as follows:

static

= −

∑

i=1

∑

j=1

i j

log

i j

(1)

where, M is the quantity of resource existing in sy-

stem S, N is the number of possible states for the i

resource and p

i j

is the probability of state j occurring

in resource i. In this context, states of resources can

be deﬁned subjectively (e.g. busy, idle and breakdown

etc.). Static complexity is essentially a measure of the

inherent complexity of a production process within

a speciﬁed time-frame, usually a year. According to

(Calinescu, 2002), static complexity reduces when re-

dundant resources are removed and product design is

simpliﬁed.

2.2 Dynamic Complexity Assessment

In contrast to static complexity, dynamic complex-

ity is a measure of a system’s operational behaviour,

especially corresponding to queues. The sources of

dynamic complexity can vary from internal to exter-

nal sources, and identifying the underlying reason for

queueing helps determine the issues in a process. The

dynamic complexity of a facility is obtained using

an expression similar to that for structural complex-

ity, except that instead of calculating probabilities for

the states as scheduled, the probabilities are estimated

from actual observations.

dynamic

= −

∑

i=1

∑

j=1

i j

log

i j

(2)

where, the p

i j

indicates probability estimates based

on observed states and N

is the number of observed

(actual) states.

Depending on whether the state of an operation

was planned or unprecedented, they can be classiﬁed

as programmable (e.g. run, set-up, idle, etc.) and

non-programmable (e.g. rework, breakdown, etc.) re-

spectively. It should be noted that non-value adding

activities such as rework, can potentially create bottle-

necks in a system. Dynamic complexity assessment in

the information theoretic approach necessitates pro-

cess observations at regular time intervals for all va-

riants available. Therefore, the information from pro-

cess should be used to determine the various parame-

ters of assessment, such as frequency and time-frame

of the measurements.

2.3 The Application of Information

Theoretic Approach

In order to assist with the understanding of the in-

formation theoretic approach, a discrete-event simu-

lation study is performed by routing two different

product variants through a battery assembly system.

The assembly system in Figure 1 produces two bat-

tery variants, of which the variant A is designed to

provide high power, whereas the variant B provides

more energy, therefore, the number and type of cells,

bus-bars, cooling system vary widely. The assem-

bly line in consideration, however, is designed to ac-

commodate both variants. The components for both

variants are available at each of the stations and the

operator is instructed to ﬁt the parts according to the

Proposing a Holistic Framework for the Assessment and Management of Manufacturing Complexity through Data-centric and

Human-centric Approaches

Figure 1: Battery assembly line layout.

work instructions. There are 7 manual stations which

are operated by 4 people. There are 3 automatic sta-

tions, which are for cell loading, ultrasonic welding

and pulse arc welding. As the variants are conveyed

through the line, an RFID reader reads the recipe pre-

sent on the pallet to determine the necessary opera-

tions that need to be performed at any station. The-

refore, the automatic stations change their conﬁgura-

tion depending on the variant in consideration. Some

of the stations do not perform any operation for the

considered variant. In this situation, the pallet is by-

passed through the station. The stations are reasona-

ble balanced to avoid potential queueing. The sig-

niﬁcant tasks that are performed manually across the

seven stations are as follows: carrier assembly, cool-

ing system install, fastening the carrier plates, plas-

tic welding, bus-bar installation, visual inspection of

welds, assembly of insulator cover. To understand the

challenges in mixed model assembly of a battery as-

sembly line with the above-mentioned characteristics,

discrete-event simulations of the system was model-

led. Various parameters were introduced and tweaked

around to visualise the ﬂow of the products in the sy-

stem.

The annual net demand for product variants are

deﬁned as 500 for each, and the available time per

year is 1920 time unit (48 time unit per week). Pro-

duct and resource information are given in Table 1

and 2, respectively. It should be noted that, variant A

is run a week, followed by running variant B for two

weeks, thus, the set-up time is calculated depending

on the next product to run. The generic states for sta-

tic complexity is determined as: set-up, make and idle

and each state is deﬁned by specifying the product as-

Table 1: Product information.

Variant A Variant B

Operations CA × (OP1A+ +OP3B+OP4B+OP5B

OP2A+OP3A+OP4A OP3B+OP4B+OP5B

+OP6A+OP7A+OP8A+ +2×OP6B+2×OP7B+

OP9A+OP10A) 2×OP8B+2×OP9B+OP10B)

Annual net demand 500 500

sociated to it (Table 2). Then, the probability of each

state is calculated (Table 3) and corresponding infor-

mation content is calculated in Table 4. Accordingly,

the static complexity of the assembly system is found

as 14.00325 bits.

Figure 2: The percentage distribution of observed states for

each workstation.

The operational complexity is analysed using dis-

COMPLEXIS 2018 - 3rd International Conference on Complexity, Future Information Systems and Risk

Table 2: Resource information (All values are given in time unit).

Stations

1 2 3 4 5 6 7 8 9 10

Operations OP1A OP1B OP2A OP2B OP3A OP3B OP4A OP4B OP5A OP6A OP6B OP7A OP7B OP8A OP8B OP9A OP9B OP10A OP10B

Set-up 0.3 0.2 0.6 0.4 0.2 0.3 0.2 0.3 0.6 0.0 0.0 0.4 0.2 0.3 0.4 0.2 0.3 0.2 0.2

Total set-up 4.0 5.4 9.6 12.8 2.4 9.6 3.2 8.0 8.8 0.0 0.0 6.4 6.4 4.8 12.8 3.2 9.6 2.4 6.4

Make 1.0 2.4 4.2 2.2 2.0 3.0 2.3 2.8 3.0 4.0 4.0 2.5 3.0 3.0 4.5 3.0 2.5 2.5 2.5

Total make 500 1200 2075 1075 1000 1500 1150 1400 1500 2000 4000.0 1250 3000 1500 4500 1500 2500 1250 2500

Capacity 1.0 2.0 2.0 2.0 1.0 4.0 3.0 4.0 3.0 2.0

Idle 210.6 322.6 658.0 633.8 411.2 420.0 490.5 402.4 573.9 36.2

Table 3: Resource/state probabilities.

Stations

1 2 3 4 5 6 7 8 9 10

Set-up A 0.002 0.005 0.001 0.002 0.005 0.000 0.003 0.003 0.002 0.001

Set-up B 0.003 0.007 0.005 0.004 0.000 0.000 0.003 0.007 0.005 0.003

Make A 0.260 0.540 0.260 0.299 0.781 0.260 0.217 0.195 0.260 0.326

Make B 0.625 0.280 0.391 0.365 0.000 0.521 0.521 0.586 0.434 0.651

Idle 0.110 0.168 0.343 0.330 0.214 0.219 0.255 0.210 0.299 0.019

Table 4: Static complexity calculation.

Stations

1 2 3 4 5 6 7 8 9 10

Set-up A 0.019 0.038 0.012 0.015 0.036 - 0.027 0.022 0.015 0.012

Set-up B 0.024 0.048 0.038 0.033 - - 0.027 0.048 0.038 0.027

Make A 0.505 0.480 0.505 0.521 0.278 0.505 0.478 0.460 0.505 0.527

Make B 0.424 0.514 0.530 0.531 - 0.490 0.490 0.452 0.523 0.403

Idle 0.350 0.432 0.529 0.528 0.476 0.480 0.503 0.472 0.521 0.108

Total (Resource) 1.322 1.513 1.615 1.628 0.790 1.475 1.526 1.454 1.602 1.078

Total (System) 14.003 bits

Figure 3: Operational complexity results of individual sta-

tions.

crete event simulation model due to the absence of ac-

tual production system data. A statistical model, e.g

normal distribution, is used to represent the demand

proﬁle of both the variants and this introduces com-

plexity which can be compared to that introduced due

to market ﬂuctuations. This signiﬁcantly increases

the time spent for set-up for each variant on proces-

sors. Moreover, the machine breakdown parameters

(a negative exponential distribution is used for mean

time between failure (MTBF) and mean time to repair

(MTTR) behaviours) are also introduced in the auto-

matic stations, namely 2, 6 and 8. Therefore, two ad-

ditional non-programmable states, which are blocked,

breakdown (including repair) are also tracked during

the model run, for which the warm up time is 200 time

units and run time is 1920 time units. Please note that,

process-ﬂow interruptions due to decision-making are

not modelled. Figure 2 shows the percentage of the

observed states per resource. The operational com-

plexity results in Figure 3 are obtained by analysing

the data gathered from the discrete event simulation

model. Accordingly, operational complexity of the

system is recorded as 16.224 bits, indicating that the

complexity occurring due to the programmable states

signiﬁcantly contributes the overall operational com-

plexity.

3 SURVEY BASED APPROACH

In this section, based on an iterative cycle consisting

of literature research and discussions with practitio-

ners, a human-centric complexity assessment metho-

dology is developed to systematically identify the root

causes of complexity within an organizational struc-

ture.

3.1 Methodology

The actual methodology of the human-centric com-

plexity assessment is divided into three main steps.

3.1.1 Step 0: Identiﬁcation of Complexity

Drivers

In order to develop the human-centric complexity as-

sessment methodology, a representative set of com-

plexity drivers were identiﬁed. Therefore, we con-

ducted an interview study and a literature review. The

survey draft was peer reviewed by experienced acade-

mics. A ﬁnal review was carried out by industry ex-

perts of a reduced sample of nine manufacturing com-

panies to improve the survey validity. Based on the

feedback the items and the format of the questionnaire

were reﬁned. As a result, a set of 20 complexity dri-

vers were deﬁned in order to evaluate the individual

complexity situation of manufacturing companies. In

the end it served as a complexity measurement tool

in a benchmarking, with 137 participating companies

from different sectors of the manufacturing industry.

Proposing a Holistic Framework for the Assessment and Management of Manufacturing Complexity through Data-centric and

Human-centric Approaches

3.1.2 Step 1: Complexity Driver Analysis

The methodology of the survey based measurement

of complexity begins with the evaluation of the com-

plexity drivers regarding their inﬂuence on the com-

pany’s complexity. The survey encompasses com-

plexity drivers which affect the whole company. In

order to estimate the perceived complexity of the spe-

ciﬁc complexity drivers within the company, the par-

ticipants are asked to rate their inﬂuence based on a

self-assessment.

3.1.3 Step 2: Calculation of Complexity Index of

Complexity Driver k

In the second step, based on the rating of the parti-

cipants a normalized complexity index is calculated.

The complexity index of each driver k is the result of

the relation of the rating of the individual complexity

driver (CD

) and the sum of the rating over all com-

plexity drivers (

∑

(3)

where:

• CI

: the relative rating of the complexity driver k,

• n: total number of complexity drivers,

• CD

: the rating of complexity driver i.

The normalization allows an evaluation which

complexity driver has the strongest inﬂuence on the

complexity within the company. Furthermore, for

evaluation purposes the complexity drivers are clus-

tered in product complexity, production complex-

ity, supply complexity, organizational complexity and

market complexity. Hence, an evaluation for each

complexity cluster can be conducted in order to de-

rive suitable methods for management.

3.1.4 Step 3: Validation of the Method

For validation purposes, the results of the assessment

tool’s complexity evaluation are discussed in a semi-

structured interview with the industry experts of the

international companies. In these interviews, an un-

derstanding is built up (based on participant percep-

tion) of the causes of complexity. Detailed discussi-

ons and explanations regarding the static and dynamic

factors reveal the challenges of the company to handle

the complexity. The results of the questionnaire and

the interview facilitate the assessment of the percei-

ved complexity of the company.

3.2 The Application of the Approach:

Precision Manufacturing (PM) Inc.

In the following, the survey based methodology is ap-

plied to the Precision Manufacturing Inc. to measure

their perceived complexity.

• Rating of the complexity drivers at PM Inc.: First,

the company rated the 19 complexity drivers on a

Likert scale (1-5). The result is shown in Table 5.

These numbers served as input for the next step in

order to normalize the complexity drivers.

• Calculation of complexity index of PM Inc.: In

this phase, each complexity driver is normalised,

as depicted in Table 5. The drivers with the big-

gest inﬂuence are product variety and program dy-

namics as well as number of customers and the

variety of customer demands.

• Validation of the complexity in a semi-structured

interview study: In a short semi-structured inter-

view, we build up an understanding of the main

complexity drivers of PM Inc. The focus of PM

Inc.’s complexity management approach is divi-

ded in product related and process related targets.

The company aims at reducing the product com-

plexity while at the same time extend their product

portfolio. Furthermore, lower production costs

and as well as a lower process variation needs to

be achieved.

Due to the high variety of customer demands and

the number of customers, PM Inc. needed to choose

suitable methods to lower the resulted high product

complexity. The activities aim at reducing the num-

ber of parts for speciﬁc existing products. By means

of modularization, the company achieved higher rates

of carry over parts and standardization within their ex-

isting product portfolio. Upcoming new products are

designed based on existing components and therefore

contribute to sustain a low number of parts.

Product variety, one of the complexity drivers with

the strongest impact on the company, also required

some changes within the assembly process. Through

modularization of the assembly process, PM Inc. ma-

naged to reduce their process variety and to establish

standardized processes. Employees designated to as-

sembly are now qualiﬁed to perform four to ﬁve pro-

ducts instead of being specialized to only one product.

Furthermore, employees from the production depart-

ment are involved in the design and approval process

of new products in order to lower the effort in the as-

sembly process.

COMPLEXIS 2018 - 3rd International Conference on Complexity, Future Information Systems and Risk

Table 5: Evaluation of complexity driver of Precision Engineering Inc.

Cluster Class Driver Rating Normalisation

1 Product complexity Product variety 5 0.8

(Product) Program dynamics 5 0.8

Product structure 3 0.05

Product technology dynamics 3 0.05

Total 16 0.06

2 Production complexity Number of value added steps 4 0.06

Production technology dynamics 3 0.05

Total 7 0.05

3 Supply chain complexity Number of suppliers 4 0.06

Uncertainty regarding delivery dates 3 0.05

Uncertainty regarding delivery quality 2 0.03

Number of distribution levels 3 0.05

Total 12 0.05

4 Organisational complexity Crosslingking degree of company processes 5 0.08

Organisational structure 4 0.06

Cooperate culture 2 0.03

Total 11 0.06

5 Market complexity Legal factors 1 0.02

Political conditions 2 0.03

Number of customers and customer groups 5 0.08

Variety of customer requirements 5 0.08

Volatility of customer demands 4 0.06

Competitive dynamics 3 0.05

Total 20 0.06

Total sum of complexity drivers 66

4 DISCUSSION

4.1 The Comparison of Two Methods

Up to this point, the authors have demonstrated the

two key sources of data for assessing complexity i.e.

objective sources such as data models/machine data

and subjective sources such as surveys and question-

naires. Furthermore, there has also been a description

as to how such sources can be transformed into tools

for assessing complexity. The data-centric approach

has been described through an example.

The data-centric complexity assessment approach

is deﬁned in this research as a means to derive com-

plexity solely through objective means. In truth, this

is not entirely possible as any manufacturing com-

plexity assessment model is ultimately designed by

humans, and the factors that would be chosen for as-

sessment are at the whim of the assessor. Neverthe-

less, the sources of the data are inherently objective.

This means that for a given assessment model, pro-

vided the relevant factors have been included, a sys-

tems complexity can be assessed and then monitored

as changes are made to control or manage it. Further-

more, data-centric approaches for complexity asses-

sment can be automated, whether by extracting data

from models e.g. CAD, virtual engineering models,

or from shop-ﬂoor machines e.g. through operati-

onal equipment effectives (OEE) calculations. This

reduces the effort of assessing complexity and there-

fore reduces the barriers associated with managing it.

In addition, such objective approaches offer compari-

sons between designs (product, process or machine),

becoming an additional metric to aid selection.

On the other hand, data-centric approaches, by de-

coupling somewhat from the human, suffer from an

inability to provide an insight into the humans per-

ception of complexity. In fact, in some cases, the

data-centric approach offers a skewed human percep-

tion of complexity as it can be expression of what the

complexity data-model designer perceives should be

included within the assessment of complexity. This

very issue can result, perhaps most critically, in a sy-

stem being deemed to be un-complex from an ob-

jective point of view but highly complex from a user

or operational point of view. One way to extract this

information is to directly ask the user. This results in a

human-centric assessment of complexity. Complexity

assessment through this lens allows one to understand

what people deem to be the sources of complexity.

This is not only a useful source of information to be-

gin the management process, but also empowers those

people asked as they are able to see how concerns rai-

sed are being addressed. Furthermore, many of the

nuances that can be picked up from an interview or a

survey are difﬁcult to obtain from data-centric appro-

aches as the latter is a model and therefore, by deﬁni-

tion, a simpliﬁcation of reality. Despite these beneﬁts,

surveys and questionnaires can be time-consuming

and therefore complex exercises and there is often

more work that needs to be done to derive an asses-

sment.

Proposing a Holistic Framework for the Assessment and Management of Manufacturing Complexity through Data-centric and

Human-centric Approaches

4.2 A Framework for the Combined

Use of Human and Data Centric

Complexity Assessment

The authors have examined the literature and to date

they have been unable to identify works that appreci-

ate the need to assess and manage complexity. The

authors therefore believe there is a signiﬁcant gap in

the literature that needs to be addressed to i) extend

the use of data models to assess complexity in in-

dustrial settings and ii) combine the use of data mo-

dels and questionnaires/surveys to assess and there-

fore manage complexity through multiple levels of

an organisation. To address this gap, a framework

is proposed (Figure 4) that combines human-centric

and data-centric complexity assessment in combina-

tion with complexity management.

The framework consists of three layers, with a sin-

gle research pillar spanning all layers. The lowest

layer of the framework is the Data Layer. This layer

represents the core data that exists in both the minds

of humans as well as that which exists in engineering

data models. This data could exist either internal or

external to the organisation as well as at almost any

level i.e. from management down to operators. As a

result of the myriad of data sources, it is necessary to

determine what data models and which people should

be interviewed in order to get the right data. For this

reason, the research pillar exists to help identify the

relevant sources or drivers of complexity. At the data

layer there exists the information necessary to assess

complexity, but at its most raw, unprocessed form. It

is necessary for some perspective to be added to the

data in order for it to be used to assess complexity.

Note that given a different perspective, a different cri-

teria concerning an enterprise could be synthesised.

Above the Data Layer there is the Assessment

Layer. At this layer the data sources are processed

so that complexity can be assessed. If the complexity

data source is a data model, then a complexity model

is used. On the other hand, if the perceptions of hu-

mans need to be used as a complexity assessor, then

the surveys or questionnaires are used on the relevant

stakeholders. At this layer again the research pillar

is fundamental in either applying or developing the

right complexity model for the relevant data source or

asking the right questions of the given stakeholder.

Based on the results of the assessment it is neces-

sary to apply the right management strategy. This is

carried out in the Management Layer. Note that at the

Data and Assessment Layers there is a distinction be-

tween the data sources and the assessment methods.

However, at the Management Layer the framework

aims to combine the complexity assessment to un-

derstand i) how they interact, and ii) to therefore im-

plement a management strategy that complements the

assessment. The research pillar is essential in deter-

mining what management strategies exist and which

are most relevant to the results derived from the com-

plexity assessment. Once the management strategy

has been identiﬁed it is implemented which closes the

loop and allows further data to be taken from the Data

Layer to understand the impact of the management

strategy. This impact is carried out through the evalu-

ation link from the Assessment layer to the Manage-

ment layer.

The framework brings together only the ele-

ments necessary to begin to combine data-centric and

human-centric complexity assessment methods. Ho-

wever, considerable work needs to be done to deter-

mine how the data should be combined i.e. the tri-

angle spanning the left and right columns in Figure 4

representing the two complexity assessment methods.

The authors’ are still investigating a robust method to

address this problem

5 CONCLUSION

This paper has described two of the key methods as-

sociated with deriving an assessment of complexity,

namely data-centric and human centric approaches.

These two classiﬁcations have not, to the best of the

knowledge of the authors, been used to categorise

complexity assessment methods previously. The re-

spective approaches are discussed and compared with

the strengths and weaknesses being identiﬁed. This

evaluation leads to the formation of an entirely no-

vel framework that combines the two assessment met-

hods into one. As well as assessing complexity, the

framework is holistic in that it appreciates the need

to manage complexity, close the loop between asses-

sment and management, and for continuously resear-

ching to identifying new complexity sources, asses-

sment methods, and management methods. The aut-

hors strongly believe that the combination of the two

assessment methods are complementary and funda-

mental to complexity management. This idea is to

be tested as part of the case study with the industrial

partner and is on-going work.

ACKNOWLEDGEMENT

The authors gratefully acknowledge the supports of

the Commission for Technology and Innovation in

Switzerland (CTI) through the project Complexity 4.0

(ref: 18993.1), the UK EPSRC through the KDCM

COMPLEXIS 2018 - 3rd International Conference on Complexity, Future Information Systems and Risk

Figure 4: Holistic Framework for combining human-centric and data-centric complexity assessment for management.

research project, and the Innovate UK through the

AMPLiFII project (ref: 102490).

REFERENCES

Alkan, B., Vera, D., Ahmad, M., Ahmad, B., and Harrison,

R. (2016a). Design evaluation of automated manufac-

turing processes based on complexity of control logic.

Procedia CIRP, 50:141–146.

Alkan, B., Vera, D., Ahmad, M., Ahmad, B., and Harri-

son, R. (2016b). A model for complexity assessment

in manual assembly operations through predetermined

motion time systems. Procedia CIRP, 44:429–434.

Alkan, B., Vera, D., Kaniappan Chinnathai, M., and Harri-

son, R. (2017). Assessing complexity of component-

based control architectures used in modular automa-

tion systems. International Journal of Computer and

Electrical Engineering, 9(1):393–402.

Budde, L., Nagler, O., and Friedli, T. (2015). A method to

set up a complexity index to improve decision-making

performance. Procedia CIRP, 36:53–58.

Calinescu, A. (2002). Manufacturing complexity: An inte-

grative information-theoretic approach. PhD thesis,

University of Oxford.

Calinescu, A., Efstathiou, J., Schirn, J., and Bermejo, J.

(1998). Applying and assessing two methods for me-

asuring complexity in manufacturing. Journal of the

Operational Research Society, 49(7):723–733.

Chinnathai, M. K., Alkan, B., and Harrison, R. (2017). Con-

vertibility evaluation of automated assembly system

designs for high variety production. Procedia CIRP,

60:74–79.

Efthymiou, K., Mourtzis, D., Pagoropoulos, A., Papakostas,

N., and Chryssolouris, G. (2016). Manufacturing sy-

stems complexity analysis methods review. Internati-

onal Journal of Computer Integrated Manufacturing,

29(9):1025–1044.

Frizelle, G. and Woodcock, E. (1995). Measuring complex-

ity as an aid to developing operational strategy. Inter-

national Journal of Operations & Production Mana-

gement, 15(5):26–39.

otzfried, M. (2013). Managing Complexity Induced

by Product Variety in Manufacturing Companies.

Complexity Evaluation and Integration in Decision-

Making. PhD thesis, University of St. Gallen.

Kohr, D., Budde, L., and Friedli, T. (2017). Identifying

complexity drivers in discrete manufacturing and pro-

cess industry. Procedia CIRP, 63:52–57.

Lindemann, U., Maurer, M., and Braun, T. (2008). Struc-

tural Complexity Management: An Approach for the

Field of Product Design.

Schuh, G. and Schwenk, U. (2001). Produktkomplexit

managen–strategien. Methoden, Tools, M

unchen,

Wien.

Sinha, K., Shougarian, N. R., and de Weck, O. L.

(2017). Complexity management for engineered sy-

stems using system value deﬁnition. In Complex Sy-

stems Design & Management, pages 155–170. Sprin-

ger.

Proposing a Holistic Framework for the Assessment and Management of Manufacturing Complexity through Data-centric and

Human-centric Approaches