Sunteți pe pagina 1din 13

Computers in Industry 105 (2019) 260–272

Contents lists available at ScienceDirect

Computers in Industry
journal homepage: www.elsevier.com/locate/compind

The design space of production planning and control for industry 4.0
Julia C. Bendula,* ,1, Henning Blunckb,2
a
Chair for the Management of Digitalization and Automation - Industry 4.0; School of Business and Economics, RWTH Aachen University, 52062 Aachen,
Germany
b
Workgroup Production & Logistics Networks, School of Mathematics and Logistics, Jacobs University, 28759 Bremen, Germany

A R T I C L E I N F O A B S T R A C T

Article history: In industrial production, distributed control is perceived to be a promising approach for dealing with
Received 28 October 2017 challenges arising from the increasing dynamical and structural complexity in that field. Currently, future
Received in revised form 18 June 2018 production systems are envisioned to be digitalized and networked systems bearing names such as
Accepted 31 October 2018
“Industry 4.0”, “Manufacturing 2.0”, “Internet of Things”, and many others. These visions share the idea of
Available online xxx
assigning tasks of production control to “intelligent” objects, such as machines, parts, and products, in
order to attain higher flexibility, higher adaptability, and therefore a higher logistics performance.
Keywords:
However, limited information and restricted computation capacity may have negative effects: The
Production system design
Factory planning
production system behavior depends on the decisions made by intelligent objects with individual and
Industry 4.0 selfish systems of objectives. This can deteriorate both the stability and the quality of achieved
Production planning and control production planning and control solutions. This trade-off situation has led researchers to the belief that a
Autonomous control combination of centralized and distributed—as well as of local and global—decision making in control
Framework might be the key to an improved and stable logistics performance. However, little is known about the
Reference architecture mechanics of these combined, semi-heterarchical control structures. Based on a profound literature
review, we compare the approaches and insights from the different research domains in order to classify
design decisions that can be made already in the planning phase of a new production system. We take a
first step towards the validation of our classification by mapping four paradigms of distributed control
into it. Our research can help the designers of future production systems to understand how to avoid the
emergence of “myopic” behavior, and it may serve as a basis for creating new control approaches by
exploring the design space.
© 2018 Elsevier B.V. All rights reserved.

1. Introduction - the need for new design paradigms for production systems to achieve higher flexibility and robustness.
industry 4.0 Productivity gains of up to 30% are expected to balance out the
disadvantages of high wages and energy costs of industrialized
In industrial production, autonomous control is perceived to be countries [2]. Terms such as “Internet of Things”, “Smart Factory”
a promising approach for achieving robustness against the and “Manufacturing 2.0” embody the vision of such highly efficient
increasing dynamical and structural complexity of production production systems, where intelligent objects, such as machines,
networks [1]. More and more data are being generated by so-called parts, and vehicles, cannot only measure and assess their own
cyber physical systems (CPS), which integrate computation with situation, but also communicate and make decisions based on local
physical objects. The increasing digitalization of production and global information (for an overview of initiatives in Europe
systems—namely, the application of advanced data processing, alone, c.f. [3]).
data storage, and data analytics, as well as cloud technologies—is The desired and reported positive effects of increased flexibility,
perceived to be the key to higher levels of automation, to more robustness, reduced complexity, and modularity are the positive
efficient processes, and to better planning and control of consequence of distributing the control functions and the decision-
making authority to multiple parallel decision-making entities,
leading to an emergent (rather than enforced) behavior of the
system [4]. However, distributed decision-making based on local
* Corresponding author. information and limited computational capacity also has negative
E-mail addresses: bendul@scm.rwth-aachen.de (J.C. Bendul),
h.blunck@jacobs-university.de (H. Blunck).
effects, since the decision-making objects cannot integrate future
1
www.scm.rwth-aachen.de. developments and other system parts into their decision-making
2
http://pln-workgroup.user.jacobs-university.de. process. As a result, the system behavior is harder to predict, and

https://doi.org/10.1016/j.compind.2018.10.010
0166-3615/© 2018 Elsevier B.V. All rights reserved.
J.C. Bendul, H. Blunck / Computers in Industry 105 (2019) 260–272 261

logistics performance may deteriorate. To limit this type of myopic towards the validation of the framework. In Section 6, we discuss
(short-sighted) behavioral traits, it is not only economically the main findings, limitations, and resulting future research
desirable, but also essential to ensure the industrial adaptation directions
of production systems with distributed control [5–7].
It has repeatedly been hypothesized that also in production 2. Control paradigms for complex production systems
systems, a balance of centralized and autonomous control yields
the best systemic performance. Moreover, it has been stated that 2.1. Digitalization of production systems – a chance to implement
the performance gain from a combination of centralized and distributed control architectures
decentralized decision strategies increases with higher levels of
decision task complexity [1,8–11]. However, new design paradigms In production research and practice, the term digitalization is
are required to ensure benefit from production systems with used to summarize a multitude of technological advancements.
intelligent objects and from a distributed control approach, for two According to [16]), digitalization is perceived among practitioners
main reasons: (1) The existence of intelligent objects and to be a combination of (1) the increased amount of data, and
distributed control is changing the role of system design itself, computational power and connectivity—including big data, open
from a situation where a designer is able to precisely predetermine data, and cloud technologies. Digitalization also gathers (2)
clearly future material flows and utilization levels c.f. [12,13], to a developments from the field of analytics and artificial intelligence,
situation where a designer only decides on the existence of such as the digitalization and automation of knowledge work by
intelligent objects (such as machines, conveyors, etc.), as well as on advanced analytics. Moreover, digitalization stands for the (3)
the rules and individual systems of objectives of these objects, and digital-physical exchange of information, for example in the form
therefore acts rather like a conductor “orchestrating” a factory to of additive manufacturing technologies, robotics, and improved
ensure that the emergent system performance is competitive and energy storage and transfer efficiency as well as (4) for improved
stable 2. The existence of intelligent objects and distributed control human-machine-interaction, in the form of touch pads, new GUIs,
is creating new trade-off situations for production system design. and virtual and augmented reality.
The designer has to decide on a certain degree and form of It is the combination of new technologies enabling a
combination between centralized and distributed control across digitalization of production (technology push) and the changing
various design decisions. In doing so, the designer accepts a certain requirements in production control, following the above dis-
degree of myopic decision-making in exchange for positive, cussed increase in complexity (application pull), that has led to
emergent system characteristics, such as higher adaptability, the development of new paradigms of production (cf. [17,18]):
quicker response, etc. Concepts such as Autonomous Control of Logistics Objects (see, for
Several design reference architectures exist for distributed instance, [1]), Internet-of-Things (cf. [19] and [20] for reviews),
production control systems which include forms of myopia control Smart Factory [21], and Networked Production [22] embody visions
(see, for example, [7]). However, these are largely experience- and of prospect organizations and of the production planning and
intuition-driven [14,15]. A better understanding of the drivers of control of manufacturing companies of the future. Due to the
and countermeasures against myopic system behavior is necessary digitalization of production, the vision of distributedly controlled—
to better navigate the trade-off between desirable and undesirable and therefore efficient—production companies and manufactur-
traits of distributed production control in smart factories. We aim ing systems, where logistics objects such as machines, parts, and
to contribute to closing this research gap at the interface of the vehicles can not only measure and assess their own situation, but
research in the fields of production system design, production can communicate with each other and make decisions based on
planning and control (PPC), and distributed control systems by local and global information, seems to be within reach for the first
proposing a research framework, seizing the design space for time. However, despite the growing interest in these distributed
production systems with distributed control architectures. Based systems and in the expected efficiency effects, no clear guidelines
on a profound literature review, including works on selfish agents, for the design and the control of such distributed architectures
game theory, and statistical physics, we present a three-fold have been provided so far [14,15,23,24]. Although all these
framework with nine dimensions which clearly indicates the concepts have linked distributed control and decentralized
parameter space that designers can work in when designing decision-making to efficiency and logistics performance gains,
production systems with distributed control architecture. the underlying cause-and-effect relationships remain unclear. So
We map four popular distributed PPC approaches—two from far, neither practice nor research has provided guidance regarding
the stream of PULL production and of agent-based manufacturing the extent to which logistics objects should be autonomous. It is
control, respectively—into the classification framework, position- not clear as to what technologies and what capabilities logistics
ing each approach along all design dimensions. We observe both objects should have, and how these objects should be connected
intra-group similarities and inter-group differences in order to with which other objects exchanging what type of information by
classify and compare distributed PPC approaches. We find that means of which systems. In terms of control hierarchy, it has to be
such mapping can show individual “fingerprints” of control clarified as to which logistics objects should take what decisions
approaches in terms of the “management” of myopic behavior. autonomously and based on what criteria and target systems.
The remainder of the article is structured as follows: In Finally, the role and the working principles of management in
Section 2, we present the state-of-the-art by outlining recent such a distributed system have to be determined in order to
developments in the digitalization of production systems and the positively influence corporate success and manufacturing perfor-
resulting growing interest in distributed control architectures, and mance.
we summarize the main findings on the different control
architectures in production systems. In Section 3, we introduce 2.2. Control paradigms for complex production systems
the concept of myopia as the guiding theoretical foundation for the
development of the classification framework of design decisions The increasing structural and dynamic complexity of produc-
with impact on the emergence of myopic behavior. In Section 4, we tion companies and manufacturing systems is affecting the
derive the design dimensions and characteristics of the classifica- planning and control effort as well as the logistics performance
tion framework. In Section 5, we use the framework to compare of manufacturing systems. Recent research literature provides only
four distributed control approaches, thus taking a first step limited insight into the situation-specific design of planning and
262 J.C. Bendul, H. Blunck / Computers in Industry 105 (2019) 260–272

control systems in order to deal with the increasing complexity of established. Therefore, an application of this knowledge in
manufacturing systems. business practice is currently not possible.
Control Theory identifies three main forms of dealing with An additional dimension was added by Trentesaux [6], who
complexity [25], but rarely provides guidelines regarding when to defined the optimal control architecture with respect to the
apply what approach in industrial practice (1) Decentralization optimization time horizon. The author states that a fully
(distribute information), (2) Decomposition (split up decision distributed control architecture is recommended for short-term
problem into smaller pieces), and (3) Model Simplification. Classical optimization, whereas a centralized control architecture is
approaches to both production planning [26] as well as organiza- recommended for long-term optimization. He concludes that for
tion structure [27] have preferred a functional decomposition, the combined achievement of short- and long-term optimality, a
leading to a hierarchy of decision-making levels, coordinated combination of hierarchical and heterarchical architectures is
through bureaucratic structures, where higher-level decisions bind necessary (Fig. 2).
lower-level decision makers [28,29]. Such hierarchical organiza-
tion structures are perceived to be best suited for stable, top-down, 3. Theoretical background - the concept of myopia and myopia
bureaucratic systems [30,31,6,32]. Changing a company’s structure control in distributed production systems
towards a more decentralized, non-hierarchical one, on the other
hand, has been identified as an important contributor towards Roughly 30 years after the idea of agent-based, heterarchical
higher levels of innovation and adaptiveness in highly dynamic production control architectures was first floated (e.g. [36]), the
environments [33,34,31,35]. industrial adoption of distributed control architectures is still lagging
The most recent wave of efforts to distribute control authority behind [14,43]. This gap has been attributed - among other factors, -
in response to increasing complexity and the ready availability of to the presence of myopic behavior [44] in distributed control
computation and sensing capabilities can be traced back to the systems and the resulting negative features, such as lack of
mid-1980s [36,37], when the mentioned researchers started to performance, unpredictability, etc. [5,6,43]. Other factors include,
discuss a complete decentralization of control authority to for instance, possible security problems and failing frequency caused
distributed decision-making entities, who were initially even by the increase in components, the advantages of central diagnostics,
barred from communication with each other [38]. commissioning and maintenance, the lack of standards, high
In production research today, it is widely acknowledged that a necessary investments, and a continuing prevalence of reductionist
combination of centralized and distributed control will give better thinking and training among production system designers and
results as compared to both strictly hierarchical and completely planners [38,45,43]; McFarlane & Bussmann 2003; [5]
distributed approaches [38,6,11]. [39]: 545f) described this in his
review of production network coordination as follows: “[ . . . ] 3.1. Triggers and effects of myopia in distributed production systems
connection and co-ordination modes should be defined balancing two
contrasting needs: (a) giving each node enough leeway and autonomy The term “myopia” has been used in management theory to
to maintain the flexibility of the network and the individual strengths describe a 30-year long debate over the (alleged) undervaluation of
of each node and (b) strict enough to render the activities of each node long-term growth perspectives by managers—in particular in Anglo-
coherent with the goals of the system.” [40]: p. 681) indicated the American countries—in favor of short-term gains (c.f. [46,47], p. 950)
need for interaction and alignment of agents in a distributed hence defines myopia as “a characteristic of a decision that overvalues
system: “The dilemma facing these networks extends to the balance short-term rewards and undervalues long term consequences”.
between having independent agents and controlling processes to meet Pressure from capital markets is the most commonly assumed
performance, which requires a strong interaction between the agents.” reason for such behavior [48]. A similar notion is also used in supply
This understanding has been transferred to production systems chain management theory, e.g. in the discussion of inventory
by several authors e.g. [1,8,41,9,10,42]. In 2007, Philipp et al. models. Here, myopic decision policies maximize the target
hypothesized that the optimal balance of centralized and function only for the next period, while ignoring the development
decentralized control regarding the logistics performance in a of planning periods further ahead [49,50].
manufacturing system is a function of system complexity and the Myopia was first used in the context of distributed PPC by
degree of distributed control (Fig. 1). However, this relationship Trentesaux [6] and said to be related to the “uncertainty of knowledge
could not yet be shown empirically, nor could the mechanistic about the future states of both the control system and the controlled
understanding of the underlying cause-effect relationship be yet system, uncertainty that increases rapidly over time” (ibid., p. 975).

Fig. 1. Optimal control architecture as a function of system complexity [119].


J.C. Bendul, H. Blunck / Computers in Industry 105 (2019) 260–272 263

Fig. 2. Optimal control architecture as a function of optimization time horizon [6].

Zambrano Rey et al. [44] state that myopic behavior appears when decision-making and the emergence of desirable and undesirable
decision-makers overemphasize current-term results at the ex- system behavior and performance characteristics are closely
pense of long-term performance. Accordingly, the authors define related. Conceptually speaking, system designers can create
myopic behavior in distributed production control systems as the systems between the two architectural poles of hierarchical PPC
inability of a decision-making entity to project the consequences of (low levels of myopia) and strictly distributed PPC (high levels of
its own decisions towards the future (temporal myopia) and myopia). To better understand and exploit the underlying cause-
towards the system as a whole (social myopia). and-effect relationships leading to the overarching hypotheses
Myopic behavior in production systems may have several that systems with combination of centralized and decentralized
reasons, including control architecture (see Section 2.2) will yield the highest logistics
performance, it is necessary to be able to characterize and explore a
 planning problem decomposition [51,52], design space between the two architectural poles.
 selfish actors [53], To implement any control system at or between these two
 decision making based on local information only (c.f. architectural poles, planners have to make several sub-decisions:
[5,6,28,54,55], and these range from questions of the form of coordination between
 bounded rationality (capacity-constraints on decision making) decision-making entities, to control network structure, the
[44]. information horizon of decision making entities, etc. In order to
integrate these various design dimensions within one conceptual
Myopic behavior can manifest itself in various forms of model, we use the concept of myopia as our theoretical foundation.
undesired system properties from increased overall cost [53], to The concept of myopia has been used before to conceptualize
system nervousness [56], to examples of chaotic [5,55] and even the negative traits of (strictly) distributed PPC and systems in
paradox system behavior (such as decreased performance under general (see Section 2.2; Zambrano Rey [9] and Zambrano Rey et al.
increased capacity, cf. [57,58]). The evolution of distributed [44]). A similar framework has been presented by Matni et al. [60]
systems in production control [38] as well as the more general for the decision-making dimensions of software-defined networks.
discussion of multi-agent systems [59] has hence fairly early seen There, the authors use (the absence of) myopia to characterize fully
the abandonment of strictly and entirely decentralized systems centralized and fully distributed myopic algorithms, which then
and has explored various approaches for controlling myopic serve as the corner stones for defining a more continuous, hybrid
behavior instead. design space for software-defined networks.
Both temporal and spatial dimensions of “near” have been used Moreover, the concept of myopia implies a reference to the
in the discussion of myopia in distributed PPC: the term “spatial properties of distributed decision-making, rather than to the
myopia” is explicitly used e.g. by Trentesaux [6], to refer to presence of computerized agents in the first place. In this, it is
situations, where “local” decision criteria (the performance of one acknowledged that also hierarchical PPC is a distributed decision-
intelligent product) guide local decision-making, instead of more making system that can be thought of and implemented as a multi-
global ones (overall schedule performance). Since both the agent system. Therefore, we can model both hierarchical and
information considered and the decision making approach taken heterarchical systems within one framework.
by individual decision-making entities is subject to a manufactur-
ing system designer’s discretion, myopic decision-making can be 4. Results - myopia control during system design and
connected to decisions of the manufacturing system designer and/ production planning and control
or planner. This assumption is shared by [44], p. 802), who note
that the “degree of myopic behavior present in heterarchical control This research aims to provide an initial framework and
architectures depends on their structural and operational design”. guidance for those researchers and practitioners aiming to design
new production systems which exploit the advantages of
3.2. Myopia as the theoretical basis for seizing the design space of distributed control architectures. The initial structure of the
production systems in the age of industry 4.0 framework goes back to the three hierarchical phases of the
initialization of new production systems determining (1) the
As shown in the previous section, the design of production (production system) design, (2) the character of (production)
systems with intelligent objects, the prevalence of myopic planning and scheduling, and (3) the character of (production)
264 J.C. Bendul, H. Blunck / Computers in Industry 105 (2019) 260–272

Fig. 3. Classification framework of design decisions. (Towards arrow tip, it is more likely to reduce myopic behavior; towards arrow base, it is more likely to increase myopic
behavior). Note that the triangular shapes are not meant to indicate an – assumed – linear relationship between the design decision and the degree of myopic behavior as the
form of this relationship is yet to be established.

control. Therefor the framework has a sequential character, and horizons, from machine level setup and staffing decisions (with the
preceding decisions influence the following decisions. The result of shortest time horizon) to factory and production network design
the considerations in this section – the proposed classification (with the longest time horizon) [63,65]. The different time scales
framework – is shown in Fig. 3. form a control hierarchy and decisions at previous steps influence
subsequent ones. Long-term PPC decisions of a strategic character
4.1. Decision-stages of production system design and production often concern production system design decisions, such as
planning and control number, location, and equipment of production facilities in order
to meet requirements regarding product mix and volume [65]. For
The term “production planning & control” (PPC) summarizes all Gudehus and Kotzab [61], one key characteristic of production
repetitive tasks of the management of the value creation processes planning and scheduling is that it often deals with the inaccurate
of a company, although the exact definitions and delineations of information of future (anticipated) demands. In the medium time-
terms vary between authors and research streams, as [61] point range, production planning is typically concerned with translating
out (see, for instance, [62–64]. PPC tasks span multiple time actual or forecasted customer demand into production orders and
J.C. Bendul, H. Blunck / Computers in Industry 105 (2019) 260–272 265

determining release dates for them [63]. In contract, production [68] describe the development of distributed manufacturing
control is typically concerned with short-term steering of released control systems as a quest to exploit flexibility potentials of a
orders ([61] [63],). system. The authors state that a distributed production system
should be designed in such a way as to keep decision alternatives
4.2. A classification of myopia-control measures for distributed open (for example, the choice of the next process step, machine,
manufacturing control final customer order, etc.) for the stage of production control,
when real-time information can be integrated into the decision-
4.2.1. Requirements and emergent system behavior as input and making process. Therefore, to create this type of decision
output variables of the classification framework alternatives, operational flexibility—namely, the ability to
The combination of system plant and system controller produce the same product mix/volume in different ways—is
constitutes the engineered and controlled system. This combina- required [69]. Operational flexibility can be measured, for
tion is set up and maintained to fulfill the formal cause of example, by the number of alternative material flow paths which
production systems—namely, to transform inflowing information result from the number and types of production processes,
and material into a specific, emergent system behavior, material- machines, routing, etc. However, flexibility itself—namely, the
ized in the form of finished products at a certain level of logistics presence of decision alternatives in a production system—can
performance—and to meet the requirements concerning the cause myopic behavior, since the decision-making entities do not
system in the form of customer demand for a certain production project consequences of their decisions onto others and therefore
portfolio [63] (Table 1). These form the input and output variables act selfishly (this form of social myopia is well-studied under the
of the classification framework presented in Fig. 3. title of “selfish routing games” and “price of anarchy”; see, for
instance, [70,71].
4.2.1.1. Design phase - plant and controller. Following the Structural complexity describes the degree to which alterna-
argumentation line of Mönch [66], we interpret production tive production paths interfere with each other cf. [72,73]. When
systems as so-called goal-seeking systems. Therefore, we planners increase the structural complexity of a production
distinguish between the design of the physical system (plant) system, the number of “second-order” effects caused by local
and the control system (controller). These types of systems can decision-making rises. Thus, it is more likely that decisions
naturally be represented as “control loops”, where any system can neglecting future consequences will negatively influence the
be decomposed into a controller and a controlled system (here, the system performance (so-called temporal myopia). For example,
plant) [67,6]. Controller and controlled system communicate Scholz-Reiter et al. [74] show that the more a production system
through the exchange of information, with sensor information structure is characterized by parallel and/or repetitive (and
flowing from the controlled system to the controller and actuator therefore less complex) elements, the more likely the emergence
signals flowing in reverse. Baker [67] specifies this concept for the of a positive impact of distributed decision making is. Jones et al.
realm of production systems where the system to be controlled is [75] state that the amount of information necessary to make good
the manufacturing plant (consisting of physical entities such as decisions in production is directly proportional to the complexity
resources and products), and the manufacturing system controller of a production system.
is the PPC system.
4.2.1.3. Controller design. In the production system design phase,
4.2.1.2. Plant design. The design of the system plant – the physical the network topology formed by decision-making entities
structure of the manufacturing system – has arguably played a connected by information flows in the production system
subdued role in the academic discussion on production planning controller is central for the emergence of myopic behavior.
and control, as the research was largely focused on developing “Intelligent” machines, parts, products, etc. are represented as
good control architectures given some physical layout (e.g. a job decision-making entities. In particular, the network topology
shop, flow-shop, etc.). In this contribution, we explicitly consider determines how these entities can access information on
both design decisions together for two reasons: (1) First, the design potentially distal parts of the system.
of the physical material flow in the production system can be Following the argumentation line of Van der Vecht et al. [4],
decisive for the direction of potential impact on the logistics Zambrano et al. [44] show that the implementation of a
performance (positive or negative) of distributed manufacturing structurally hierarchical agent network topology is a feasible
control. (2) Second, the physical system design frames the decision approach for improving coordination and reducing myopia in a
options the PPC system has. A control system for a flexible job shop production system. Similarly, Helbing et al. [76] show theoretically
has a considerably larger option space as compared to that of a that hierarchical networks have the fastest information transmis-
strictly linear flow production, where planning decisions are sion across networks, and Kearns et al. [77] experimentally show
mostly reduced to the sequence of products. By discussing physical that in communication networks, the existence of agents with high
and control system design together, we acknowledge and reflect connectivity can enhance system performance.
that the need to control myopic decision making is affected by the While we consider strictly hierarchical and heterarchical
option space provided by the physical system design. system architectures as the extreme ends of our scale, the variety
Within the plant design phase, there are two decisions with in the middle may include concepts such as autonomous clusters
impact on the emergence of myopic behavior: (1) operational [72], heterarchies with shortcuts, cross-links in hierarchies, other
flexibility and (2) structural complexity. forms of agent topologies [78,79].

Table 1
Input and output variables of the classification framework.

Variables Short definition Measures (selection)


requirements (anticipated/past) customer demand (fluctuations) for a specific number of products, product architecture (number of parts, construction
product portfolio principle, etc.),
emergent system production output (quantities, fluctuations quality, logistics productivity, lead time, due date reliability, inventory levels
behavior performance)
266 J.C. Bendul, H. Blunck / Computers in Industry 105 (2019) 260–272

Table 2
Decision dimensions and scales on the design level.

Dimension Short definition Scale (high to low)


operational flexibility The ability to produce the same product mix/volume in different ways in one many to no path alternatives
production system
structural complexity The degree to which alternative production paths interfere with each other. high to low complexity
network topology architecture of the network formed by intelligent decision-making entities completely heterarchical to completely hierarchical
linked by information or material flows

Table 2 gives an overview of the decision dimensions on the to assess the “restrictiveness” of a schedule. Three similar
design level. The decision dimensions form the upper part of the measures that aim to express the degree of schedule flexibility
classification framework presented in Fig. 3. as the ratio of possible production sequences with and without
some offline schedule are discussed by Artigues et al. [90,91], and
4.2.1.4. Scheduling. Scheduling is one of the most relevant Blunck & Bendul [84].
decision-making tasks taking place before the release of the
production schedule [80]. Schneeweiß [28] points out that the 4.2.1.5. Production control. In the production control phase, three
strong information asymmetry between scheduling and schedule main decisions influence the emergence of myopic behavior,
execution (e.g., caused by missing and/or outdated information namely the temporal information horizon, the approach to
during scheduling, such as machine failures, actual processing nervousness alleviation, and the competition approach.
times, etc.) requires that schedule deviations are either reacted to Temporal information horizon stands for the amount of time
or anticipated during scheduling [81]. for which information on the expected future evolution of the
A purely distributed approach would arguably apply a purely system or on the past system behavior is available to the decision-
reactive scheduling approach (postponing all sequencing and making unit. In its most simple (and myopically inclined) form,
allocation decisions to the time of their execution). decentralized control architectures will make decisions solely on
Partial schedules have been presented as one solution approach information about the current state of some subset of system
for controlling myopic behavior: partial schedules pre- elements. A prominent example for this type of control are
determine the solution to a subset of scheduling decisions. In dispatching rules, such as the queue length estimator (QLE) (cf.
scheduling theory, several authors have shown that pre-selecting [92]). Making information about future system states available
operations may eliminate the so-called “Graham Anomalies” (see may reduce the emergence of temporal myopia (for instance, in the
[57]) as the loss of production system performance (see for form of requirements forecasting in MRP II systems). Approaches
instance [82]; Artigues, et al. [83]). Using partial schedules to that use past information to predict future behavior include
control myopic behavior in distributed manufacturing control biology-inspired approaches, such as stigmergy-based approaches,
systems was first discussed in (Bongaerts, et al 2000). However, which resemble communication processes of insects, such as ants
it is not clear how the pre-selective policies described above can cf. [93–95], and bees cf [96]. Where decision-making entities are
be applied to flexible manufacturing systems. First work has able to abstract between a planning and an execution level, both
been presented by Blunck & Bendul [84], who show that a partial online planning and offline planning can be performed by either
schedule generated by randomly deleting information from a drawing information from the current system state and passing it
flexible job-shop schedule does in fact deteriorate system on to a central simulation entity (cf. [97]) or by computing decision
performance. alternatives locally (inducing an increased amount of inter-agent
Also in the research stream of selfish routing games, partial communication, cf [98,99].). The range of planning approaches for
schedules have been discussed. It was shown that prescribing the evaluating available information ranges from simulation to
route choice for a subset of traffic (for instance, solving the optimization models (e.g. [100]).
assignment problem), may reduce the emergence of myopic In such continuous (re-)evaluation of decision alternatives, the
behavior [85–87], though not for arbitrary networks [88]. The level of system nervousness describes the extent to which decision-
results of static scheduling do not have to be (unlike so far making entities tend to immediately follow up the change in the
assumed) used for explicit coordination (i.e. making the agent at observed system state with a change in their own decision making
runtime adhere to a previously made decision). It can also be [56]. Setting a higher level of nervousness alleviation may avoid
applied in order to merely direct computational efforts during the emergence of chaotic system behavior arising from entities
runtime toward promising regions of the solution space, by who respond to every change in the observed system state by
favoring investigation of solutions close to the static scheduling changing their own decisions. In this context, the discouraging (or
solution (cf [24].). blocking) of repeated decision changes of agents is an established
Table 3 gives an overview of the decision dimension on the strategy to calm a system [56,101] and to improve system
planning and scheduling level forming the medium part of the convergence [102,103].
classification framework presented in Fig. 3. The competition approach determines how decision-making
This design dimension stands out in the context of the entities make decisions with regard to their peers. In agent-theory
framework proposed here, in that there exist methods to quantify in particular, a range of possible behaviors for agents is discussed,
the degree to which a schedule constraints the set of possible PPC ranging from a competitive / antagonistic character on the one
decisions that were permissible in a given system design. The extreme to an altruistic approach on the other extreme (e.g.
reviewer is pointed to [89], Appendix A.1) for a review of measures [104,105]). The reader is referred to Blunck & Bendul [106] for

Table 3
Decision dimensions and scales on the planning and scheduling level.

Dimension Short definition Scale (high to low)


Schedule flexibility The degree to which the scheduling is already completed before production Fully flexible to fully scheduled
J.C. Bendul, H. Blunck / Computers in Industry 105 (2019) 260–272 267

Table 4
Decision dimensions and scales on the control level.

Dimension Short definition Scale (high to low)


temporal information the amount of time over which (expected) system behavior information on current state to covering past and/or containing plan for
horizon is known future
nervousness alleviation the extent to which decision-making entities follow none to strong
changes
in the observed system state
competition approach approach decision-making entities follow to make competitive / antagonistic to cooperative to altruistic
decisions

further references. Table 4 gives an overview of the decision 5.2. ConWIP control
dimensions on the control level. The decision dimensions form the
lower part of the classification framework presented in Fig. 3. Compared to kanban systems, so-called constant work in
progress (ConWIP) control can be applied in more diverse material
5. Mapping archetypes of control architectures into the flow layouts and can be implemented in environments with less
framework stable demand and with wider product portfolios [109,110].
However, the main application field is still the material supply
This section validates the above-developed classification and at of production lines.
the same time highlights one possible way in which application of Despite the similarity in physical layout, ConWIP has a
the classification model may help researchers and practitioners in hierarchical network topology, because the pull principle is only
the domain of manufacturing system design. We have selected four applied to the first and last workstations of the system, whereas
control approaches, two so-called PULL approaches and two information exchange is organized centrally [63,108]. The central
approaches that are based in the field of holonic manufacturing order-release list pre-determines the allocation of orders as well as
systems (HMS), with elements of distributed control in order to most sequencing decisions. Thus, ConWIP is characterized by a
map them into the developed framework. PULL production rather low level of schedule flexibility. In a ConWIP controlled
systems have, as part of lean manufacturing systems, arguably system, the general system topology is based on historical data,
been the most widely applied form of distributed production whereas order release is based on forecast data [63]). As in kanban
control to date. The two examples of HMS were chosen due to their systems, no system nervousness alleviation methods are necessary
wide application and good coverage of the requirements posed on for ConWIP control, because system entities take no autonomous
agent-based manufacturing control systems [7]. Since the goal of reactions towards environmental changes. The decision-making
our research is to derive a framework for the comparison and entities in a ConWIP control system generally ignore other decision
evaluation of control approaches, we have decided for already makers' target functions and are thus of a competitive character.
well-established, mature control approaches to point out the
general potential of our approach. Moreover, we have decided to 5.3. PROSA reference architecture
choose two related approaches coming from two different research
streams in order to show that the framework derived is also Production systems with the so-called product-resource-order-
capable of indicating intra- and intergroup similarities and staff-architecture (PROSA) ([42,98,111],) are flexible systems
differences. without strong restrictions on physical layout. The PROSA
Through the mapping, we can both compare the control approach is based on the main idea that in every production
approaches regarding the design towards myopia emergence and system, three main problems have to be dealt with: resource
take a first step towards the validation of the derived framework. In utilization, technology aspects to achieve the quality demanded,
this, the framework can serve as the basis for the development of and logistics performance regarding delivery time and due dates.
new control architectures. Based on this understanding, PROSA makes use of three holon
types - resource holons (similar to the intelligent agents for
5.1. Kanban systems entities, such as machines, conveyors, pallets, parts etc.), product
holons (including the necessary process and product knowledge,
In terms of physical plant design, kanban systems are such as product life cycle, bill of materials, work plan etc.) and
characterized by low levels of flexibility and complexity [4]. order holons (with control of the production order). The order
They can be applied to material flow layouts of clear scope holon is responsible for the accomplishment of all necessary
(typically two workstations) for systems with stable demand and logistics and production processes and decisions regarding
a very limited number of product variants. The resulting control planning, scheduling, and resource allocation. The different holon
network topology of a kanban system is of a heterarchical types exchange information on the production system and the
character, mainly based on historical data, and the information actual product-, resource- and order-related states in order to
exchange between decision makers on the same hierarchy levels control the system [6,24,15,98,111].
is very limited [107]. Moreover, kanban systems also have little The mentioned authors generally present PROSA as a control
reliance on scheduling, as they aim for a stable, leveled approach which is suitable for job shop environments. Job shop
production environment [108]. systems are characterized not only by their high flexibility in terms
In kanban systems, information is exchanged solely via kanban of physical layouts, but—typically for job shop systems—by a high
cards. The recent demand and the information horizon of such number of product variants that can be created through the high
systems are very short [109]. Such systems possess few nervous number of alternative production paths in the network. At the
alleviation measures, since they assume stable production same time, this type of system can be labeled as rather complex,
environments. Finally, the participants of the production systems since production orders may interfere with each other.
should be characterized as competitive, because they do not PROSA architectures are characterized by a mix of heterarchical
consider the benefit of other players or workstations. and hierarchical information structures due to the coexistence of
268 J.C. Bendul, H. Blunck / Computers in Industry 105 (2019) 260–272

so-called holarchies and independent holons [24,42]. Verstraete & design space that have been subject to less academic interest and
Valckenaers, [111] present a method to include a predefined may hence deserve future attention, such as increased efforts to
production plan into the PROSA holons’ consideration. Although study the impact of more altruistic agent behavior (c.f. e.g. [115]).
PROSA systems are capable of following plans in general, we
consider them to be only slightly bound by partial scheduling. 6.2. Similarities and differences of the four control approaches
Verstraete et al. [24] show that in a PROSA-controlled system,
information about the immediate future can be made available. In The mapping of the four control approaches onto the
terms of nervousness alleviation, Verstraete et al. [98] propose a classification framework brings to light similarities and differences
solution of oppressing decision changes. Holon agents can be between the four control approaches.
characterized as selfish and competitive when fulfilling their Comparing the two PULL approaches already indicates how
targets without considering the benefits of other agents. aiming at more complex production environments requires
increased investment in other forms of myopia control (here:
5.4. ADACOR reference architecture hierarchical system structure and an extended temporal informa-
tion horizon). We may understand the transition from PULL
The ADAptive holonic COntrol aRchitecture for distributed principles towards agent-based production control architectures
manufacturing systems (ADACOR) reference architecture was as a continuation of this effort to bring distributed control to
proposed in [112]. It was created for volatile production systems increasingly complex production environments: Comparing PULL
with frequent changes and disturbances. The focus is on flexible principles with HMS reference architectures, we see a clear shift
manufacturing systems in the form of job shops with high numbers towards aiming at more complex production systems, demanding
of variants and high numbers of path alternatives through the the implementation of more sophisticated myopia control
system, and therefore with high complexity. ADACOR aims to be as approaches within scheduling and control. In this context, it is
decentralized as possible and as centralized as necessary. In other worthwhile noting that the visions of future production systems
words, a centralized approach is chosen for optimization purposes, (such as “Industry 4.0”, “Manufacturing 2.0”, and “Internet of
and a decentralized approach is chosen when unexpected events Things”), typically refer to low volume, high variability types
and modifications are likely [112]. Four holon types are defined, of manufacturing environments. However, so far, distributed
namely product, task, operational, and supervisor holons [112]. production control has only been applied successfully in
In this aspect, ADACOR control systems are more hierarchical industrial practice in the form of kanban and ConWIP control
than PROSA systems. ADACOR aims to establish stable advisory in production systems with little to no flexibility and low levels of
hierarchies and only switches from this stationary to a more complexity.
decentralized reactive control architecture as disturbances occur, The two PULL approaches and the two HMS- or agent-based
while PROSA can be understood as a fluid approach to adapting to manufacturing control approaches span opposite ends of the
new environments [24,42,113]. In addition, ADACOR systems do design space in terms of flexibility and complexity. Thus, we can
not emphasize peer-to-peer communication between operation argue that agent-based control approaches may be understood as
holons on the same hierarchy level. Similarly to PROSA systems, a an attempt to transfer the benefits from distributed control—that
production schedule is not necessary for an ADACOR system to have proven themselves so far in simple layouts and mass
function, as the plans are generated within the system by production environments—to job shops with small production
supervisor holons. As no method of integrating master production quantities.
plans is mentioned for the ADACOR systems, we assume the The two PULL approaches, kanban and ConWIP, differ mainly in
reference model to make no restrictions on schedule flexibility. terms of their scheduling flexibility. However, the mapping shows
According to [113], the temporal information horizon of that both PULL approaches Kanban and ConWIP control the
ADACOR agents covers at least the short-term future, so that the negative consequences of myopic behavior. This has its foundation
agents may predict the duration and impact of disturbances. in the explicitly modelled agents with decision-making authority
ADACOR systems aim to shorten the time of transient periods and is therefore not traceable in the “control” dimensions of the
when it comes to reacting to system changes. Therefore, classification framework.
nervousness alleviation is lower than it is in PROSA systems, Here, the classification framework reflects the development in
which proscribe changes without significant benefits [113,24]. research: ConWIP control can serve as an example for the case that
ADACOR agents can be understood as selfish competitors, since the (re-)introducing concepts of hierarchical control may be necessary
benefits of other agents remain unconsidered when making their to benefit from distributed control in more complex production
decisions. settings.
The comparison of the two HMS approaches shows that the
6. Discussion, limitations & future research development from the PROSA approach to the (more recent)
ADACOR approach has brought the focus onto nervousness
6.1. Model validation alleviation strategies.
Across the four frameworks considered here, none places an
The mapping of the four control approaches into the emphasis on changing the competition approach of the agents.
classification framework is a first step towards the latter’s Promising results of initial experiments in the domain of HMS (e.g.
validation. We can show that (1) all four different distributed [44]) are yet to proliferate into widely used frameworks.
control approaches show differentiable patterns (“fingerprints”)
when mapped to the classification framework and (2) related 6.3. Limitations and future research directions
control architectures (here: the two PULL and the two HMS
approaches) show higher intra-group similarity than compared to There are four main limitations to the classification framework
members of the other group. In this, we may argue that the presented: (1) So far, the model neglects interrelationships
classification framework provides sufficient accuracy in the between the design dimensions. The design dimensions for
intended application context and may hence be considered distributed systems are not independent, but may require,
validated [114]. A more elaborate mapping of further reference support, or impair each other (already noted by [116]). This
models and their derivatives should further highlight parts of the was most prominently noted in the discussion on “hierarchy”,
J.C. Bendul, H. Blunck / Computers in Industry 105 (2019) 260–272 269

implying that internal consistency of the model cannot be determine whether a specific control architecture is appropriate for
ensured, since not all relationships between concepts have solving a given manufacturing system control problem?” [23], p.
explicitly been considered [117]. (2) The manufacturing system 881). However, qualitatively seizing the design space may serve as
requirements (here included as an “input” to the classification a prerequisite for further operationalization. (4) The triangular
framework), should be more specific. The design dimensions are shapes used in Figs. 3 and 4 to indicate higher and lower levels of
discussed with regard to their impact on myopic behavior, but the myopia may be misinterpreted as linear relations between the
degree of myopic behavior which is tolerable—even desirable—in design decision and myopic behavior. However, this research is
a given production system setting, has to be contingent on the only a first step to better understand the emergence and effects of
requirements posed on that system. (3) So far, the design myopia in general. Given the lack of quantification of both the
dimensions have not been operationalized. The aim to analyti- design dimensions (as the independent variable) as well as the
cally understand interdependencies between the traits of the degree of myopic behavior (the dependent variable) means that
problem description and the (likely) best control approach has the form of relationship between the two cannot be determined as
been of high interest in production research for many years. For of now. Eventually, a sustained and enduring push by the academic
instance, in 1997, Rogers and Brennan put this question in the community to address the limitations mentioned above should
spotlight: “Of primary importance is the fundamental question allow us to develop a clearer picture of the nature of this
concerning the choice of control architecture: i.e. is it possible to relationships.

Fig. 4. Mapping the reviewed control approach into the classification framework.
270 J.C. Bendul, H. Blunck / Computers in Industry 105 (2019) 260–272

7. Conclusion [15] H. Van Dyke Parunak, Agents in overalls: experiences and issues in the
development and deployment of industrial agent-based systems, Int. J. Coop.
Inf. Syst. 09.03 (2000) 209–227.
In this paper, we have reviewed literature from multiple [16] McKinsey and Company, Industry 4.0 | How to Navigate Digitization of the
research domains and the various theoretical findings and design Manufacturing Sector., McKinsey & Company, 2015.
approaches reported therein, to create a classification of design [17] Heiner Lasi, Peter Fettke, Hans-Georg Kemper, Thomas Feld, Michael
Hoffmann, Industry 4.0, Bus. Inf. Syst. Eng. 6.4 (2014) 239–242.
decisions, through which the emergence of myopic behavior in a [18] Rajkumar, Ragunathan, Insup Lee, Lui Sha, and John Stankovic (2010).
distributed manufacturing control system can be controlled. Cyberphysical Systems: The Next Computing Revolution Proceedings of the
While the literature provides several examples of and evidence 47th Design Automation Conference, DAC’ 10, Anaheim California. New York,
NY, USA: ACM, pp. 731-736.
for each mechanism, the generalizability, scaling behavior, and [19] Luigi Atzori, Antonio Iera, Giacomo Morabito, The internet of things: a survey,
interplay of the design decisions are yet unknown, and a Computer Networks, vol. 54.15(2010) , pp. 2787–2805.
profound mechanistic understanding of the interplay between [20] Jayavardhana Gubbi, Rajkumar Buyya, Slaven Marusic, Marimuthu
Palaniswami, Internet of things (IoT): a vision, architectural elements, and
system design decisions and the emergence of myopic behavior
future directions, Future Generation Computer Systems 29.7. Including
is still not available. We have demonstrated how the classifica- Special sections: Cyber-enabled Distributed Computing for Ubiquitous Cloud
tion model can be applied in order to differentiate between and Network Services & Cloud Computing and Scientific Applications | Big
existing control approaches based on distributed control Data, Scalable Analytics, and Beyond, (2013) , pp. 1645–1660.
[21] Detlef Zuehlke, Smart factory | towards a factory-of-things, Annu. Rev.
architectures. This type of classification model can be used for Control 34.1 (2010) 129–138.
the development of new control architectures that are both (1) [22] The Economist Intelligence Unit, Networked Manufacturing: The Digital
distinct from existing approaches, i.e. by exploring a so far Future. Briefing Paper, The Economist Intelligence Unit Ltd., 2014.
[23] Paul Rogers, Robert W. Brennan, A simulation testbed for comparing the
untouched region of the design space, and that both (2) exhibit performance of alternative control architectures, in: S. Andradorrit, K.J. Healy, D.
basic features for reducing myopic behavior. Our paper is meant H. Withers, B.L. Nelson (Eds.), Proceedings of the 29th Winter Simulation
to provide initial proof of concept, based on the definitions of the Conference, IEEE Computer Society, Atlanta, GA, USA, 1997, pp. 880–887.
[24] Verstraete, Paul, Bart Saint Germain, Paul Valckenaers, Hendrik Van Brussel,
architectures reviewed. Jan Belle, Hadeli Hadeli, Engineering manufacturing control systems using
PROSA and delegate MAS, Int. J. Agent-oriented Softw. Eng. 2.1 (2008) 62–89.
References [25] Lubomír Bakule, Decentralized control: an overview, Annual Reviews in
Control, 32.1(2008) , pp. 87–98.
[26] Arnoldo C. Hax, Harlan C. Meal, Hierarchical Integration of Production
[1] Bernd Scholz-Reiter, Michael Görges, Thorsten Philipp, Autonomously
Planning and Scheduling. Sloan School of Management Working Paper 656-
controlled production systems | Influence of autonomous control level on
73, Massachusetts Institute of Technology., Cambridge, MA, 1973.
logistic performance, CIRP Ann. Manuf. Technol. 58.1 (2009) 395–398.
[27] Thomas W. Malone, Stephen A. Smith, Modeling the performance of
[2] Wilhelm Bauer, Sebastian Schlund, Dirk Marrenbach, Oliver Ganschar,
organizational structures, Oper. Res. 36.3 (1988) 421–436.
Industrie 4.0 - Volkswirtschaftliches Potenzial für Deutschland. Studie.
[28] Christoph Schneeweiß, Distributed Decision Making, 2nd, Springer, Berlin,
BITKOM Bundesverband Informationswirtschaft, Telekommunikation und
Heidelberg, New York, 2003.
neue Medien e.V. and Fraunhofer Institut für Arbeitswirtschaft und
[29] Thomas W. Malone, Kevin Crowston, The interdisciplinary study of
Organisation (IAO), 2014.
coordination, ACM Comput. Surv. 26.1 (Mar. 1994) (1994) 87–119.
[3] European Commission, Digitising European Industry Accessed: 2017-04-11.
[30] Mary Uhl-Bien, Russ Marion, Bill McKelvey, Complexity Leadership Theory:
url:, (2016) . https://ec.europa.eu/digital-single-market/en/digitising-
shifting leadership from the industrial age to the knowledge era. The
european-industry.
Leadership Quarterly 18.4. Leadership and Complexity, (2007) , pp. 298–318.
[4] Bob Van der Vecht, Frank Dignum, John-Jules Ch. Meyer, Martijn Neef, A
[31] Tom E. Burns, G.M. Stalker, The Management of Innovation”. University of
dynamic coordination mechanism using adjustable autonomy.
Illinois at Urbana-Champaign’s Academy for Entrepreneurial Leadership
Coordination, organizations, institutions, and norms in agent systems III,
Historical Research Reference in Entrepreneurship Available at SSRN:, (1961)
in: Jaime Simão Sichman, Julian Padget, Sascha Ossowski, Pablo Noriega
. https://ssrn.com/abstract=1496187.
(Eds.), Lecture Notes in Computer Science, vol. 4870, Springer, Berlin,
[32] Benyamin B. Lichtenstein, Mary Uhl-Bien, Russ Marion, Anson Seers, James
Heidelberg, 2008, pp. 83–96.
Douglas Orton, Craig Schreiber, Complexity leadership theory: an interactive
[5] Weiming Shen, Lihui Wang, Qi Hao, Agent-based distributed manufacturing
perspective on leading in complex adaptive systems, Emerg. Complex. Organ.
process planning and scheduling: a state-of-the-art survey. Systems, man,
8.4 (2006) 2–12.
and cybernetics, part c: applications and reviews, IEEE Trans. 36.4 (2006)
[33] Ralph D. Stacey, The science of complexity: an alternative perspective for
563–577.
strategic change processes, Strateg. Manag. J. 16.6 (1995) 477–495.
[6] Damien Trentesaux, Distributed Control of Production Systems. Engineering
[34] David L. Levy, Applications and limitations of complexity theory in
Applications of Artificial Intelligence 22.7. Distributed Control of Production
organization theory and strategy, in: Jack Rabin, Gerald L. Miller, W.
Systems, (2009) , pp. 971–978.
Bartley Hildreth 2nd (Eds.), Handbook of Strategic Management, Marcel
[7] Adriana Giret, Damien Trentesaux, Software engineering method for
Dekker, Inc., New York, Basel, 2000, pp. 67–87.
intelligent manufacturing systems: a comparative survey. Industrial
[35] L. Monostori, J. Váncza, S.R.T. Kumara, Agent-based systems for
applications of holonic and multi-agent systems, in: Vladimir Marík, Arnd
manufacturing, CIRP Ann. Manuf. Technol. 55.2 (2006) 697–720.
Schirrmann, Damien Trentesaux, Pavel Vrba (Eds.), Lecture Notes in
[36] J.ózsef Hatvany, Intelligence and cooperation in heterarchic manufacturing
Computer Science, vol. 9266, Springer International Publishing, 2015, pp. 11–
systems, Robot. Comput. Manuf. 2.2 (1985) 101–104.
21.
[37] Neil A. Duffie, Rex S. Piper, Nonhierarchical control of manufacturing
[8] T. Philipp, F. Böse, K. Windt, Evaluation of autonomously controlled logistic
systems, J. Manuf. Syst. 5.2 (1986) 141.
processes, in: Roberto Teti (Ed.), Proceedings of 5th CIRP International
[38] D.M. Dilts, N.P. Boyd, H.H. Whorms, The evolution of control architectures for
Seminar on Intelligent Computation in Manufacturing Engineering,
automated manufacturing systems, J. Manuf. Syst. 10 (1) (1991) 79–93.
University of Naples, Ischia, Italy, 2006, pp. 347–352.
[39] Guido Nassimbeni, Network structures and co-ordination mechanisms: a
[9] Zambrano Rey, Thérése Bonte Gabriel, Vittaldas Prabhu, Damien Trentesaux,
taxonomy, Int. J. Oper. Prod. Manag. 18.6 (1998) 538–554.
Reducing Myopic Behavior in FMS Control: a Semi-heterarchical Simulation-
[40] David Bennett, Rob Dekkers, Industrial networks of the future: a critical
optimization Approach. Simulation Modelling Practice and Theory 46.0.
commentary on research and practice, in: Krisztina Demeter (Ed.), Papers of
Simulation-optimization of Complex Systems: Methods and Applications,
the 12th International EurOMA Conference on Operational and Global
(2014) , pp. 53–75.
Competitiveness, 2005, pp. 677–686.
[10] Thomas Y. Choi, Kevin J. Dooley, Manus Rungtusanatham, Supply networks
[41] Katja Windt, Felix Böse, Thorsten Philipp, Autonomy in production logistics:
and complex adaptive systems: control versus emergence, J. Oper. Manag.
identification, characterisation and application, Robotics and Computer-
19.3 (2001) 351–366.
Integrated Manufacturing 24.4. ICMR2005: Third International Conference
[11] Luc Bongaerts, Laszlo Monostori, Duncan McFarlane, Botond Káadár,
on Manufacturing Research, (2008) , pp. 572–578.
Hierarchy in distributed shop floor control, Computers in Industry, vol.
[42] Hendrik Van Brussel, Jo Wyns, Paul Valckenaers, Luc Bongaerts, Patrick
43.2(2000) , pp. 123–137.
Peeters, Reference architecture for holonic manufacturing systems: PROSA,
[12] Michael Schenk, Siegfried Wirth, Egon Müller, Factory Planning Manual -
Comput. Ind. 37.3 (1998) 255–274.
Situation-driven Production Facility Planning. 1st, Springer Science +
[43] Vladimír Marík, Duncan McFarlane, Industrial adoption of agent-based
Business Media, Berlin, Heidelberg, 2010.
technologies, Intell. Syst. IEEE 20.1 (2005) 27–35.
[13] George Chryssolouris, in: Frederick F. Ling, William Howard Hart 2nd (Eds.),
[44] Zambrano Rey, Cyrille Pach Gabriel, Nassima Aissani, Abdelghani Bekrar,
Manufacturing Systems: Theory and Practice, Springer-Verlag, 2006.
Thierry Berger, Damien Trentesaux, The control of myopic behavior in semi-
[14] A.M. Farid, L. Ribeiro, An Axiomatic Design of a Multi-agent Reconfigurable
heterarchical production systems: a holonic framework, Eng. Appl. Artif.
Mechatronic System Architecture. Industrial Informatics, IEEE Transactions
Intell. 26.2 (2013) 800–817.
on 11.5 (Oct. 2015), (2015) , pp. 1142–1155.
J.C. Bendul, H. Blunck / Computers in Industry 105 (2019) 260–272 271

[45] V. Marík, J. Lazansky, Industrial applications of agents technologies, Control its Applications 363.1, Information and Material Flows in Complex Networks,
Eng. Pract. 15 (2007) 1364–1380. 2006, pp. 141–150.
[46] Alan Felstead, Tracing the connections: short-termism, training and [77] Michael Kearns, Siddharth Suri, Nick Montfort, An experimental study of the
recession, Int. J. Hum. Resour. Manag. (2016). coloring problem on human subject networks, Science 313.5788 (2006) 824–827.
[47] Kevin J. Laverty, Managerial myopia or systemic short-termism? The [78] Bryan Horling, Victor Lesser, A survey of multi-agent organizational
importance of managerial systems in valuing the long term, Manag. Decis. paradigms, Knowl. Eng. Rev. 19.4 (Dec. 2004) (2004) 281–316.
42.8 (2004) 949–962. [79] David Isern, David Sánchez, Antonio Moreno, Organizational structures
[48] David Marginson, Laurie McAulay, Exploring the debate on short termism: a supported by agent-oriented methodologies, J. Syst. Softw. 84.2 (2011) 169–
theoretical and empirical analysis, Strateg. Manag. J. 29.3 (2008) 273–292. 184.
[49] William S. Lovejoy, Stopped myopic policies in some inventory models with [80] Vincent T’Kindt, Jean-Charles Billaut, Multicriteria Scheduling |Theory,
generalized demand processes, Manag. Sci. 38.5 (1992) 688–707. Models and Algorithms, Springer, Berlin, 2006.
[50] Yunzeng Wang, The optimality of myopic stocking policies for systems with [81] Djamila Ouelhadj, Sanja Petrovic, A survey of dynamic scheduling in
decreasing purchasing prices, Eur. J. Oper. Res. 133.1 (2001) 153–159. manufacturing systems, J. Sched. 12.4 (2009) 417–431.
[51] A. Tharumarajah, Survey of resource allocation methods for distributed [82] Jean-Claude Billaut, François Roubellat, Characterization of a set of schedules
manufacturing systems, Prod. Plan. Control. 12.1 (2001) 58–68. in a multiple resource context, J. Decis. Syst. 5.1-2 (1996) 95–109.
[52] Yves Pochet, Laurence A. Wolsey, Production Planning by Mixed Integer [83] S. Wu, Eui-Seok Byeon David, Robert H. Storer, A graph-theoretic
Programming. Springer Series in Operations Research and Financial decomposition of the job shop scheduling problem to achieve scheduling
Engineering, Springer, New York, 2006. robustness, Oper. Res. 47.1 (1999) 113–124.
[53] Pradeep Dubey, Inefficiency of nash equilibria, Math. Oper. Res. 11.1 (1986) 1– [84] Henning Blunck, Julia Bendul, Invariant-based production control reviewed:
8. mixing hierarchical and heterarchical control in flexible job shop
[54] Grace Yuh-Jiun Lin, James J. Solberg, Effectiveness of flexible routing control, environments. Industrial applications of holonic and multi-agent systems,
Int. J. Flex. Manuf. Syst. 3 (3-4) (1991) 189–211. in: Vladimír Marík, Arnd Schirrmann, Damien Trentesaux, Pavel Vrba (Eds.),
[55] Paulo Leitão, Agent-based distributed manufacturing control: a state of-the- Lecture Notes in Computer Science, vol. 9266, Springer International
art survey, Eng. Appl. Artif. Intell. 22.7 (2009) 979–991 Distributed Control of Publishing, 2015, pp. 96–107.
Production Systems,. [85] Tim Roughgarden, Stackelberg scheduling strategies, SIAM J. Comput. 33.2
[56] Karuna Hadeli, Paul Verstraete, Bart Saint Germain, Hendrik Van Brussel, A (2004) 332–350.
study of system nervousness in multi-agent manufacturing control system. [86] V.S.Anil Kumar, Madhav V. Marathe, Improved results for stackelberg
Engineering self-organising systems, in: Sven A. Brueckner, Giovanna Di scheduling strategies, Peter Widmayer, Stephan Eidenbenz, Francisco
Marzo Serugendo, David Hales, Franco Zambonelli (Eds.), Lecture Notes in Triguero, Rafael Morales, Ricardo Conejo, Matthew Hennessy (Eds.),
Computer Science, Vol. 3910, Springer, Berlin, Heidelberg, 2006, pp. 232–243. Automata, Languages and Programming: 29th International Colloquium,
[57] R.L. Graham, Bounds for certain multiprocessing anomalies, Bell Syst. ICALP 2002 Málaga, Spain, July 8-13, 2002 Proceedings (2002) 776–787.
Technol. J. 45.9 (1966) 1563–1581. [87] Yannis A. Korilis, Aurel A. Lazar, Ariel Orda, Achieving network optima using
[58] Dietrich Braess, Über ein Paradoxon aus der Verkehrsplanung, stackelberg routing strategies, IEEE. ACM 5.1 (Feb. 1997) (1997) 161–173.
Unternehmensforschung, vol. 12.1(1968) , pp. 258–268. [88] Vincenzo Bonifaci, Tobias Harks, Guido Schäfer, Stackelberg routing in
[59] W. Alshabi, S. Ramaswamy, M. Itmi, H. Abdulrab, Coordination, cooperation arbitrary networks, Mathematics of Operations Research, vol. 35.2(2010) , pp.
and conflict resolution in multi-agent systems, in: Tarek Sobh (Ed.), 330–346.
Innovations and Advanced Techniques in Computer and Information [89] Rainer Kolisch, Make-to-Order Assembly Management, 1st ed., Springer
Sciences and Engineering, Springer, Netherlands, 2007, pp. 495–500. Science + Business Media, 2001.
[60] Nikolai Matni, Ao Tang, John C. Doyle, A case study in network architecture [90] Christian Artigues, Jean-Charles Billaut, Carl Esswein, Maximization of
tradeoffs, Proceedings of the 1st ACM SIGCOMM Symposium on Software solution flexibility for robust shop scheduling, Eur. J. Oper. Res. (2005) 314–
Defined Networking Research. SOSR’ 15, ACM, Santa Clara, California, 2015 328 165.2. Project Management and Scheduling.
18:1-18:7. [91] Aloulou Mohamed Ali, Marie-Claude Portmann, “An efficient proactive-
[61] Timm Gudehus, Herbert Kotzab, Comprehensive Logistics, Springer, Berlin, reactive scheduling approach to Hedge against shop floor disturbances.”
2009. English, in: Graham Kendall, Edmund K. Burke, Sanja Petro-vic, Michel
[62] S. Gershwin, R. Hildebrant, Rajan Suri, S. Mitter, A Control Perspective on Gendreau (Eds.), Multidisciplinary Scheduling: Theory and Applications,
Recent Trends in Manufacturing Systems, vol. 6.2, IEEE Control Systems Springer, 2005, pp. 223–246.
Magazine, 1986, pp. 3–15 (Apr. 1986). [92] Joc Cing Tay, Nhu Binh Ho, Evolving dispatching rules using genetic
[63] Hopp, Wallace J and Mark L Spearman (2008). Factory physics. 3rd. McGraw- programming for solving multi-objective flexible job-shop problems,
Hill/Irwin New York. isbn: 978-007-123246-3. Comput. Ind. Eng. 54.3 (2008) 453–473.
[64] G.ünther Schuh (Ed.), Produktionsplanung und -steuerung. 3rd, Springer, [93] Patrick Peeters, Hendrik Van Brussel, Paul Valckenaers, Jo Wyns, Luc
Berlin, Heidelberg, 2006. Bongaerts, Martin Kollingbaum, Tapio Heikkilá, Pheromone based emergent
[65] Harish C. Bahl, Larry P. Ritzman, Jatinder N.D. Gupta, Determining Lot Sizes shop floor control system for flexible flow shops, Artificial Intelligence in
and Resource Requirements: A Review. Operations Research, vol. 35.3(1987) , Engineering 15.4. Methodology of Emergent Synthesis, (2001) , pp. 343–352.
pp. 329–345. [94] H. Van Dyke Parunak, Go to the ant: engineering principles from natural
[66] Lars Mönch, Agentenbasierte Produktionssteuerung komplexer multi-agent systems, Ann. Oper. Res. 75.0 (1997) 69–101.
Produktionssysteme. 1st, Deutscher Universitäts Verlag, Wiesbaden, 2005. [95] D. Armbruster, C. de Beer, M. Freitag, T. Jagalski, C. Ringhofer, Autonomous
[67] Albert D. Baker, A survey of factory control algorithms that can be control of production networks using a pheromone approach, Physica a:
implemented in a multi-agent heterarchy: dispatching, scheduling, and Statistical Mechanics and Its Applications 363.1, Information and Material
pull, J. Manuf. Syst. 17.4 (1998) 297–320. Flows in Complex Networks, 2006, pp. 104–114.
[68] Robert W. Brennan, Douglas H. Norrie, Metrics for evaluating distributed [96] Bernd Scholz-Reiter, Thomas Jagalski, Julia C. Bendul, Autonomous control of
manufacturing control systems, Computers in Industry 51.2, Virtual a shop floor based on Bee’s foraging behaviour, in: Hans-J.örg Kreowski,
Enterprise Management, 2003, pp. 225–235. Bernd Scholz-Reiter, Hans-Dietrich Haasis (Eds.), Dynamics in Logistics,
[69] George Chryssolouris, Konstantinos Efthymiou, Nikolaos Papakostas, Springer, Berlin, Heidelberg, 2008, pp. 415–423.
Dimitris Mourtzis, Aris Pagoropoulos, Flexibility and complexity: is it a [97] O. Cardin, P. Castagna, Using online simulation in holonic manufacturing
trade-off? Int. J. Prod. Res. 51 (23-24) (2013) 6788–6802. systems, Eng. Appl. Artif. Intell. 22 (7) (2009) 1025–1033.
[70] Tim Roughgarden, Eva Tardos, How bad is selfish routing? J. ACM 49.2 (Mar. [98] Verstraete, Paul, Paul Valckenaers, Bart Saint Germain, Hendrik Van Brussel,
2002) (2002) 236–259. Karuna Hedeli, Integration of planning systems and an agent-oriented MES,
[71] Tim Roughgarden, The price of anarchy is independent of the network Int. J. Manuf. Technol. Manag. 8 (1-3) (2006) 159–174.
topology, J. Comput. Syst. Sci. 67.2 (2003) 341–364 Special Issue on STOC [99] Tom Holvoet, Danny Weyns, Paul Valckenaers, Patterns of delegate MAS". In:
2002. self-adaptive and self-organizing systems, 2009. SASO’ 09, Third IEEE
[72] Rok Vrabi9 c, Damir Husejnagi9c, Peter Butala, Discovering autonomous International Conference on, (2009) , pp. 1–9.
structures within complex networks of work systems, CIRP Ann. Manuf. [100] S.S. Heragu, R.J. Graves, Byung-In Kim, A. St Onge, Intelligent agent based
Technol. 61.1 (2012) 423–426. framework for manufacturing systems control, Systems, Man and
[73] Gerry Frizelle, Eric Woodcock, Measuring complexity as an aid to developing Cybernetics, Part A: Systems and Humans, IEEE Transactions on 32.5 (Sept.
operational strategy, Int. J. Oper. Prod. Manag. 15.5 (1995) 26–39. 2002) (2002) 560–573.
[74] B. Scholz-Reiter, M. Freitag, Ch. de Beer, Th. Jagalski, Modelling dynamics of [101] H. Van Dyke Parunak, Sven A. Brueckner, Robert Matthews, John Sauter, How
Autonomous logistic processes: discrete-event versus continuous to calm hyperactive agents, Proceedings of the Second International Joint
approaches, CIRP Annals - Manufacturing Technology, vol. 54.1(2005) , pp. Conference on Autonomous Agents and Multiagent Systems. AAMAS’ 03,
413–416. ACM, Melbourne, Australia, 2003, pp. 1092–1093.
[75] Albert T. Jones, Larry H. Reeker, Abhijit V. Deshmukh, On information and [102] Borislav Hadzhiev, Katja Windt, Werner Bergholz, Marc-Thorsten Hütt, A
performance of complex manufacturing systems, Proceedings of the model of graph coloring dynamics with attention waves and strategic
Manufacturing Complexity Network Conference. (2002) 2002. waiting, Adv. Complex Syst. 12.6 (2009) 549–564.
[76] Dirk Helbing, Hendrik Ammoser, Christian Kühnert, Information flows in [103] Marta Sales-Pardo, Daniel Diermeier, Luís A.Nunes Amaral, The impact of
hierarchical networks and the capability of organizations to successfully individual biases on consensus formation, PLoS One 8.5 (May 2013) (2013)
respond to failures, crises, and disasters, Physica A: Statistical Mechanics and e58989.
272 J.C. Bendul, H. Blunck / Computers in Industry 105 (2019) 260–272

[104] Nicholas R. Jennings, J.R. Campos, Towards a social level characterization of [113] Paulo Leitão, Armando W. Colombo, Francisco Restivo, An approach to the
socially responsible agents, Software Engineering. IEE Proceedings (1997) formal specification of Holonic control systems, Vladimír Marík, Duncan
11–25. McFarlane, Paul Valckenaers (Eds.), Holonic and Multi-Agent Systems for
[105] Shaw Green, Leon Hurst, Brenda Nangle, P.ádraig Cunningham, Fergal Manufacturing: First International Conference on Industrial Applications of
Somers, Richard Evans, Software Agents: a Review, Tech. rep. Trinity College Holonic and Multi-Agent Systems, HoloMAS 2003, Prague, Czech Republic,
Dublin, Department of Computer Science, 1997. September 1st-3rd, 2003 (2003) 59–70.
[106] Henning Blunck, Julia Bendul, Controlling myopic behavior in distributed [114] Stewart Schlesinger, Roy E. Crosbie, Roland E. Gagné, George S. Innis, C.S.
production systems classification of design choices, Procedia CIRP 57. Lalwani, Joseph Loch, Richard J. Sylvester, Richard D. Wright, Naim Kheir,
Supplement C. Factories of the Future in the Digital Environment - Dale Bartos, Terminology for model credibility, Simulation 32.3 (1979) 103–
Proceedings of the 49th CIRP Conference on Manufacturing Systems, (2016) , 104.
pp. 158–163. [115] Sascha Ossowski, Ana García-Serrano, Social structure in artificial agent
[107] Mark L. Spearman, Michael A. Zazanis, Push and pull production systems: societies: implications for Autonomous problem-solving agents. Intelligent
issues and comparisons, Oper. Res. 40.3 (1992) 521–532. agents V: agents theories, architectures, and languages, in: J.örg P. Müller,
[108] Johnny C. Ho, Yih-Long Chang, An integrated MRP and JIT framework, Anand S. Rao, Munindar P. Singh (Eds.), Lecture Notes in Computer Science,
Comput. Ind. Eng. 41.2 (2001) 173–185. Vol. 1555, Springer, Berlin, Heidelberg, 1999, pp. 133–148.
[109] Mark L. Spearman, David L. Woodruff, Wallace J. Hopp, CONWIP: a pull [116] Les Gasser, Social conceptions of knowledge and action: DAI foundations and
alternative to kanban, Int. J. Prod. Res. 28.5 (1990) 879–894. open systems semantics, Artif. Intell. 47.1 (1991) 107–138.
[110] Sarah M. Ryan, Bruno Baynat, F. Fred Choobineh, Determining inventory levels [117] John G. Wacker, A definition of theory: research guidelines for different
in a CONWIP controlled job shop, IIE Trans. 32.2 (2000) 105–114. theory-building research methods in operations management, J. Oper.
[111] Paul Verstraete, Paul Valckenaers, Towards cooperating planning and Manag. 16.4 (1998) 361–385.
manufacturing execution systems, IFAC Proceedings Volumes 39.3. [119] Thorsten Philipp, Christoph Beer, Katja Windt, Bernd Scholz-Reiter,
Proceedings of the 12th IFAC Symposium on Information Control Problems in Evaluation of autonomous logistic processes- analysis of the influence of
Manufacturing, (2006) , pp. 393–398. structural complexity, in: Michael Hülsmann, Katja Windt (Eds.),
[112] Paulo Leitão, Francisco Restivo, ADACOR: a holonic architecture for agile and Understanding Autonomous Cooperation and Control in Logistics, Springer,
adaptive manufacturing control, Comput. Ind. 57.2 (2006) 121–130. Berlin, Heidelberg, 2007, pp. 303–324.

S-ar putea să vă placă și