Overview of the PhD Project: Agile Control Architecture for

Reconﬁgurable Manufacturing Systems

Bringing Flexible Manufacturing to the Next Level

Dani

el Telgen

, Erik Puik

, Leo van Moergestel

and John-Jules Meyer

Research Centre Technology & Innovation, HU University of Applied Sciences Utrecht,

Nijenoord 1, 3552 AS Utrecht, The Netherlands

Dept. of Information and Computing Sciences, Utrecht University,

Princetonplein 5, 3584 CC Utrecht, The Netherland

1 INTRODUCTION

The manufacturing industry is changing. This change

is driven by several developments, including techno-

logical changes, customer expectations, and paradigm

shifts in the manufacturing industry itself. Most no-

table developments in technological changes are 3D

printing, which can be applied to different ﬁelds, in-

cluding metal printing and the possibility to quickly

produce prototypes. This gives the possibility to

quickly automate the manufacturing of shapes that

were originally manually crafted. Considering the

paradigm shift, there are two large changes which can

be seen from the business and engineering perspec-

tive. The business is organizing their company pro-

cesses to be able to quickly adapt to market changes,

providing quick product to market abilities, in man-

ufacturing this is called ’Agile Manufacturing’ (Gu-

nasekaran, 1999). From a (mechanical) engineering

perspective machine builders are designing their ma-

chine to be modular, with the goal to standardize parts

so they can easily be replaced, reused, or reconﬁgured

(ElMaraghy, 2005). The change can also be seen from

the customer point of view. While once you only had

one choice of the color of a car, you can now choose

your own color, bluetooth kit and wheelset.

With 3D printers the diversity and opportunities to

create speciﬁc parts will likely increase dramatically.

This could be a huge boost to creativity, where people

can print their own glasses, or spare parts for their

classic restored car. This might create new markets

for automated mass customization.

2 RESEARCH OVERVIEW

In this PhD project the goal is to research the use

and implementation of new technologies to create a

shorter time to market for automated manufacturing

of new products. This should enable the possibility to

especially automate the manufacturing of ’high mix,

low volume’ products that could originally only be

done (cost-efﬁciently) by manual labor. To make this

possible, a new way of manufacturing has been in-

troduced, called ’Grid Manufacturing’ (Puik and Mo-

ergestel, 2010). Grids use a set of reconﬁgurable ma-

chine systems, called ’equiplets’ to offer a diversity

of generic services. In this concept an equiplet is a

simple, autonomous low cost platform that is highly

standardized so it can easily be reconﬁgured to offer

new manufacturing capabilities. Hence, equiplets are

able to quickly adapt to new market needs.

The concept of Reconﬁgurable Machine Systems

(RMS) (Koren and Shpitalni, 2010), like equiplets,

have been researched before. However, focus has

mainly been on the modularity of the hardware. In the

ﬁeld of Agent Technology, software has been used to

create ﬂexible logistics, that are also required for grid

manufacturing. This introduces problems whereof

some have been widely researched, like the job-shop

problem. However, the use of agent technology on

a lower level, like hardware control and direct in-

dustrial use has been very limited. Hence in this

PhD project it was chosen to take a bottom up ap-

proach where autonomous systems will be used to

create ﬂexible manufacturing systems for equiplets

and grids. This quickly introduces many fundamen-

tal and practical problems, e.g., capability descrip-

tions of the hardware, inﬂuence of reconﬁguration to

Supervisory Control and Data Acquisition (SCADA)

systems. This introduces new varieties to schedul-

ing problems. Instead of optimizing scheduling, re-

sources could be changed or added to improve efﬁ-

ciency.

The research project is built around the concept of

grids and equiplets. In this concept all systems have a

virtual counterpart, i.e., agents to represent them. Ba-

sically the manufacturing process is completely dy-

Telgen D., Puik E., van Moergestel L. and Meyer J..

Overview of the PhD Project: Agile Control Architecture for Reconﬁgurable Manufacturing Systems - Bringing Flexible Manufacturing to the Next Level.

 2014 SCITEPRESS (Science and Technology Publications, Lda.)

A product requests a list of equiplets for every manufacturing step

that needs to be completed. The system automatically translate

the requested steps towards the capabilities (services) that the

equiplets offer.

Once all the feasable equiplets have been

found, the product agent will start the

scheduling process. During this process it will

create a schedule based on which equiplet can

handle the most (consecutive) product steps,

and has the least load. After an equiplet has

been chosen, the equiplet will be contacted in

order to schedule.

A product is defined by a collection of product steps.

These products steps are abstract descriptions of

manufacturing steps. Product steps can be processed

by equiplets which offer the required service.

Services divide product steps in service steps, which

are more specific actions. The service steps are

divided in equiplet steps by the modules. Equiplet

steps are concrete steps (which can be directly

interpreterd to machine instructions) which ROS can

perform.

Whenever the product is ready to be produced, part

that are needed for production will have to be

transported to the equiplets. The product agent is

responsible for transporting the parts to the proper

equiplet. Parts will be transported on a single robot

that is scheduled by the equiplet, capable of carrying

all parts needed for one complete product. This

robot will navigate through the grid autonomously.

After the product and required parts are

available at the right equiplet, the product

steps will be executed by the equiplets. The

responsible product agent will receive

feedback from the equiplet containing

information about the executed product step.

This information will remain with the product

agent for diverse purposes. Including recycling

processes or possible repairs.

Figure 1: Basic representation of one step in the manufacturing process.

namic. Figure 1 shows the basic process of how one

step in the manufacturing process is performed.

3 OUTLINE OF OBJECTIVES

The objectives are to create a fully functional grid us-

ing a limited amount of equiplets. These will be used

to answer the following research questions:

• Which technologies can be used to enable ﬂexible

manufacturing?

• What are the properties of a hybrid platform and

are these properties compliant for industrial use?

• Can the intelligent behavior of autonomous sys-

tems be compliant with current or future industrial

safety regulations?

• How can you describe a product of which its man-

ufacturing process is independent of speciﬁc man-

ufacturing hardware?

• Is it possible to change (add/remove/reconﬁgure)

manufacturing resources in an active manufactur-

ing environment? Which effects will this have on

the overall system.

• Which technologies can be applied effectively

to minimize the installation and setup times for

the software of reconﬁgured manufacturing ma-

chines?

• Is it possible to effectively use simulation / emu-

lation to plan reconﬁguration actions in the future

that improve the overall efﬁciency of a grid.

These research questions are, as mentioned before

in the introduction, meant to achieve a shorter time

to market. This is achieved by providing the ability

to cost-efﬁciently use automated manufacturing for

medium product quantities. Speciﬁc main objectives

are the creation of a hybrid architecture with both the

performance, safety and stability characteristics with

industrial speciﬁcation. Secondly, systems should be

autonomous, so disturbances within the grid will lead

to minimum problems to equiplets. The equiplets

themselves should be reconﬁgurable with a number of

modules that can automatically be reconﬁgured. The

grid itself should also be reconﬁgurable by adding or

removing entire equiplets with minimum disturbance

to other systems in the (partly active) grid.

Secondary objectives are the research of schedul-

ing and optimization routines for the grid. This might

involve a study of distributed (heterarchical) versus

centralized (hierarchical) scheduling and investiga-

tion of the general business case.

Besides these objectives other researchers related

to this project have speciﬁc objectives including the

ICAART2014-DoctoralConsortium

research of cost efﬁciency of grids (Puik et al., 2011),

product/process modeling of reconﬁgurable manufac-

turing systems (Puik et al., 2013) and several aspects

of the product agents, including scheduling, recycling

and repair (van Moergestel et al., 2013).

4 STAGE OF THE RESEARCH

Currently most of the project has been focused on

the implementation of several prototype equiplets that

can be used to create the grid. This involves both

hardware and software developments; since this is a

multi disciplinary endeavor, literature is still being

studied in parallel with the development process. The

results of this study will be used for the experimental

design of the deliberative systems that are part of the

the grid software architecture.

The hardware of several prototype equiplets have

been completed, with a diversity of hardware mod-

ules. Including a 3D printer module and a deltarobot

for pick and place actions. Figure 2 shows a proto-

type equiplet platform and model, conﬁgured with a

deltarobot pick and place module for assembly pur-

poses.

To be able to conduct research a proof of con-

cept hybrid software architecture has been created

that combines both a reactive and deliberative layer

using agent technology. This platform is being used

to research several essential topics, including:

• The Hybrid architecture based on Robot Operat-

ing System (ROS) for hardware control and Multi

Agent Systems (MAS) using Java Agent DEvel-

opment Framework (JADE)

• Automatic Translation from a product description

to hardware instructions

• Reconﬁguration of equiplets

• System behavior and safety of reconﬁgurable ma-

chine systems

• Resource Management in a (changing) Grid

• Grid Simulation and optimization

The ﬁrst topic, hybrid architecture is nearing com-

pletion; this was based on a study of the requirements

which have been published (Telgen et al., 2013a). As

a result of this the architecture shown in Figure 3 was

developed. This platform is divided in three layers.

The Modules layer controls the hardware on the lower

level. Modules are sensors or actuators which require

speciﬁc software. On the equiplet level the modules

are controlled by an equiplet node. Nodes are sepa-

rate processes in the Robot Operating System, a spe-

ciﬁc software framework for robot software develop-

Figure 2: On the left of an equiplet with a deltarobot and

glassplate working area module attached. Right a real

equiplet prototype with a deltraobot module and a glass-

plate working area that is used to observe with a computer

vision system to localize parts.

ment (Quigley et al., 2009). On an equiplet level the

equiplet node takes commands from a blackboard that

is controlled by an equiplet agent. On the ROS level

there are also some smarter processes, like computer

vision. These are conducted on the ROS level to en-

sure their performance. Data that is perceived by the

sensors is modeled in the environment node. The grid

level encompasses all systems that are common for all

equiplets. It has several databases which can be used

to acquire speciﬁc software for new conﬁgurations

and to store logistics information. Product agents are

mainly set on the grid layer to negotiate with multiple

equiplets, depending on the required manufacturing

steps that they need to perform. The process of the

negotiation has also been studied using simulations

and published (van Moergestel et al., 2012).

5 RESEARCH PROBLEM

The main problem of the research project is how to

efﬁciently create reconﬁgurable systems that can be,

ﬂexible, autonomous, safe, efﬁcient, cooperative, pre-

dictable, low cost, fast, and easy to use. The ﬁrst prob-

lem in this task is to discover how to combine the re-

active layer that controls all hardware with a delibera-

tive layer that controls all higher processes. This will

guarantee both the performance required for hardware

control and more intelligent behavior that is necessary

for the decision processes on higher layers, that have

to deal with the ﬂexible nature of grids.

OverviewofthePhDProject:AgileControlArchitectureforReconfigurableManufacturingSystems-BringingFlexible

ManufacturingtotheNextLevel

Figure 3: Overview of the basic hybrid architecture using Robot Operating System and a Multi Agent System based on JADE.

Once a proof of concept architecture has been de-

signed, the next problem will be to make the systems

reconﬁgurable. This creates several problems, since

one of the goals is that reconﬁguring hardware will

not require any new software coding and/or compil-

ing. To make this possible the systems should be

easily conﬁgurable and be able to automatically load

new software modules, like plug and play devices in a

PC. However, in contrast with PC’s, machines are less

standardized and have actuators that can also have a

high dependency on each other, e.g., a gripper placed

on a robotic arm changes the working area of the ac-

tuator, which has an impact on safety behavior. Also

some actuators require physical calibration. As such a

large amount of knowledge is required to support the

automatic reconﬁguration of the software for recon-

ﬁgurable machines.

Once reconﬁguration is possible, capability and

resource management become important. Products

need to know which steps can be reconﬁgured at

which equiplet. Since equiplets can be reconﬁgured,

the manufaturing resources can be changed during

runtime. This gives a variation on the job-shop prob-

lem, which in its static form is already NP hard. How-

ever, in this case the resources (shops) can be changed

to optimize the manufacturing (jobs) in the system.

Besides these main problems, there are also sev-

eral practical and interesting problems, including how

to dynamically ﬁnd objects (using computer vision),

logistics (connecting to ERP systems and transport

logistics between equiplets), scaling and stability is-

sues. As far as feasible within the time constraints,

these problems will be taken into account.

6 STATE OF THE ART

Several aspects can be considered state of the art. To

discuss this matter, this chapter shall give an overview

of comparable work and the differences with the cur-

rent research project.

Paulo Leit

a has published an extensive survey of

agent technology in manufacturing in 2009, he clearly

states in his conclusion that traditional manufacturing

control systems are adapted case by case. Hence, it

is expensive and time consuming to develop, main-

tain and reconﬁgure. A large aspect of this is the lim-

ited ability to deal with disturbances of any kind. As

such, he stresses that traditional manufacturing con-

trol needs to be renewed with emerging technologies

to create agile and reconﬁgurable architectures. In

his survey he also mentions that control systems in

the context of reconﬁgurable manufacturing systems

are extremely rare and usually restricted in its imple-

ICAART2014-DoctoralConsortium

mented functionalities. The adoption of these tech-

nologies in industry are therefore very limited (Leit

ao,

2009). Because of this it is important that in the cur-

rent research project real functionality can be shown

to industry to prove its effectiveness. This had led to

the actual development of prototype systems that are

shown to industry

(Heintz et al., 2007) and (J

arvenp

a and Torvi-

nen, 2013) discuss some topics which are related in

this research. Heintz et al. discusses the sense-

reasoning gap, applied to UAV’s where abstract in-

formation is step by step deliberated to speciﬁc infor-

mation. J

arvenp

a shows how to match product re-

quirements with system capabilities of the manufac-

turing equipment. In our research we add both these

principles and take it further to use the capabilities of

the machines to automatically translate these to spe-

ciﬁc instructions that control the hardware that will

be used. This translation process will be conducted

in real time and does not require compilation of code.

Instructions are communicated directly by the agents

to the lower ROS level (Telgen et al., 2013b).

Schild and Busmann are well known for their

work in using agents in manufacturing. Especially

the 2000+ production system

is of interest. This

system was able to manufacture a number of varia-

tions of cylinder heads and has been in active use at a

Mercedes-Benz engine plant. In this system it was

possible to change some programs for each variant

and to use a ﬂexible transportation system. This sys-

tem utilized Computerized Numeric Control (CNC)

machines. While this system was successful, it was

also expensive and limited to CNC’s machines that

could automatically switch tools. So far, the 2000+

system seems to be closest to our goals, it was de-

commissioned in 2005 after the product’s life cycle

was ended. No new systems were setup, mainly be-

cause of the high costs and difﬁcult business case.

In contrast with 2000+, our research project fo-

cuses on even more ﬂexibility, to such extend that ma-

chines can offer a wide variety of products within a

grid. Basically this is achieved by lowering the cost

of standardized platforms and make them offer one

single simple service. These services have a large di-

versity, since the machines offer generic (non product

speciﬁc) services. Including computer vision cam-

era and assembly operations. Reconﬁguration aspects

therefore become more important, since there are a

Demos of the current project are shown at the

precision fair every year: http://www.precisiebeurs.nl/

intro-en-us/ last visited December 2013

2000+ production system: http://www.stefan-

bussmann.de/en/agents/p2000p.html last visited December

2013

wide range of capabilities. This offers a range of

problems as mentioned in the research problem chap-

ter.

While these referrals show how to place this

project as state of the art in comparison with other

work there are also general changes in the market that

enable this research. For example the need for mass

customer-unique products, that is fueled by 3D printer

technology, and the return of high-tech manufacturing

to the western countries are an important stimulant for

this research project.

7 METHODOLOGY

This research will be conducted at an applied univer-

sity. As such the practical feasibility and therefore

valorisation at the industry is considered to be extra

important. Hence, an entire grid will be completed,

including hardware and software which will be used

as a proof of concept for a grid using equiplets. This

inﬂuences the research, since it becomes harder to iso-

late individual aspects of the work, because all results

have to be placed in the larger context of industrial

ﬂexible manufacturing and can not be limited to a

simulated model. At all times all results should be

validated in a practical context of the entire system.

Hence, this project is given an experimental character

that has to be validated with empirical data. While

this is possible for some research aspects, the large

scope of the project might make it hard to use quan-

titative analysis of all aspects. As a result, the scope

will need to be made smaller by deciding on some

aspects using qualitative methods. An example of

this is how the basic platform for the software of the

equiplets was chosen (Telgen et al., 2013a).

8 EXPECTED OUTCOME

On a practical level the main goals are to make a small

grid with a number of reconﬁgurable equiplets which

can demonstrate the proposed technologies. This in-

cludes reconﬁguration, safety aspects and the auto-

mated scheduling and instruction generation. This

proof of concept will be analyzed to proof its effec-

tives to industry. Formalization of the agents, ontolo-

gies and the automated translation are most interest-

ing from a scientiﬁc perspective. However, the prac-

tical appliance and testing of the hybrid architecture

are also considered essential to prove how these kind

of technologies can be valorized.

Besides the physical creation of a small grid there

is the development of a full simulated grid, where

OverviewofthePhDProject:AgileControlArchitectureforReconfigurableManufacturingSystems-BringingFlexible

ManufacturingtotheNextLevel

hardware will be emulated to be used in a fully func-

tional MAS that controls all higher level aspects.

These can be used to conduct several experiments like

optimization through the use of reconﬁguration and

other logistical aspects like error behavior, etc.

9 CONCLUSIONS

This article gives an overview of the topics in this of

the PhD project named: ’An Agile Control Architec-

ture for Reconﬁgurable Manufacturing Systems’. It

introduces the concepts of equiplets and grid man-

ufacturing and shortly describes the involved prob-

lems and goals. Some aspects, like general concepts,

the hybrid architecture and automatic translations of

manufacturing steps to instructions have already been

published. However, many aspects still require more

research. Especially the metrics on the full archi-

tecture will provide more insight in the effectiveness

of this approach. A big challenge is using the rela-

tively new technologies and prove their suitability for

real industrial use. However, if this can be (partly)

achieved this could potentially have a high impact on

industry.

On of the risks of this project is the large scope.

Speciﬁc research questions are limited by the large

amount of possible implementations. The implemen-

tations have then to be tested in a complete live sys-

tem, running several safety and practical limitations

that might inﬂuence the metrics. While this is a chal-

lenge, it is expected that this research will be useful

for new research and industrial projects in the future.

REFERENCES

ElMaraghy, H. (2005). Flexible and reconﬁgurable manu-

facturing systems paradigms. International Journal of

Flexible Manufacturing Systems, 17(4):261–276.

Gunasekaran, A. (1999). Agile manufacturing: a frame-

work for research and development. International

journal of production economics, 62(1):87–105.

Heintz, F., Rudol, P., and Doherty, P. (2007). Bridging

the sense-reasoning gap using dyknow: A knowledge

processing middleware framework. In Hertzberg,

J., Beetz, M., and Englert, R., editors, KI 2007:

Advances in Artiﬁcial Intelligence, volume 4667 of

Lecture Notes in Computer Science, pages 460–463.

Springer Berlin Heidelberg.

arvenp

a, E. and Torvinen, S. (2013). Capability-based

approach for evaluating the impact of product require-

ment changes on the production system. In Azevedo,

A., editor, Advances in Sustainable and Competitive

Manufacturing Systems, Lecture Notes in Mechanical

Engineering, pages 173–185. Springer International

Publishing.

Koren, Y. and Shpitalni, M. (2010). Design of reconﬁg-

urable manufacturing systems. Journal of Manufac-

turing Systems, 29(4):130 – 141.

Leit

ao, P. (2009). Agent-based distributed manufacturing

control: A state-of-the-art survey. Eng. Appl. Artif.

Intell., 22(7):979–991.

Puik, E. and Moergestel, L. (2010). Agile multi-parallel

micro manufacturing using a grid of equiplets. In

Ratchev, S., editor, Precision Assembly Technologies

and Systems, volume 315 of IFIP Advances in Infor-

mation and Communication Technology, pages 271–

282. Springer Berlin Heidelberg.

Puik, E., Telgen, D., Moergestel, L., and Ceglarek, D.

(2013). Structured analysis of reconﬁgurable manu-

facturing systems. In Azevedo, A., editor, Advances in

Sustainable and Competitive Manufacturing Systems,

Lecture Notes in Mechanical Engineering, pages 147–

157. Springer International Publishing.

Puik, E., van Moergestel, L., and Telgen, D. (2011). Cost

modelling for micro manufacturing logistics when us-

ing a grid of equiplets. In Assembly and Manufactur-

ing (ISAM), 2011 IEEE International Symposium on,

pages 1–4.

Quigley, M., Conley, K., Gerkey, B. P., Faust, J., Foote, T.,

Leibs, J., Wheeler, R., and Ng, A. Y. (2009). Ros: an

open-source robot operating system. In ICRA Work-

shop on Open Source Software.

Telgen, D., van Moergestel, L., Puik, E., Muller, P., and

Meyer, J.-J. (2013a). Requirements and matching

software technologies for sustainable and agile man-

ufacturing systems. In INTELLI 2013, The Second

International Conference on Intelligent Systems and

Applications, pages 30–35.

Telgen, D., van Moergestel, L., Puik, E., van Zanten, A.,

Abdulamir, A., and Meyer, J.-J. (2013b). Agile prod-

uct manufacturing by dynamically generating control

instructions. In Assembly and Manufacturing (ISAM),

2013 IEEE International Symposium on, pages 282–

284. IEEE.

van Moergestel, L., Langerak, W., Meerstra, G., van

Nieuwenburg, N., Pape, F., Telgen, D., Puik, E., and

Meyer, J.-J. (2013). Agents in domestic environments.

In Control Systems and Computer Science (CSCS),

2013 19th International Conference on, pages 487–

494. IEEE.

van Moergestel, L., Puik, E., Telgen, D., and Meyer, J.-J.

(2012). Production scheduling in an agile agent-based

production grid. In Web Intelligence and Intelligent

Agent Technology (WI-IAT), 2012 IEEE/WIC/ACM

International Conferences on, volume 2, pages 293–

298.

ICAART2014-DoctoralConsortium