This chapter discusses the fundamental notion of structure for an organization, typically a company. There are many structures that implicate the way complexity is tackled in organizations, hence the need to distinguish them successfully.
The discussion provided below comes as an extension of Mintzberg’s theory of organizations [MIN 96] and multi-agent systems principles in a networked enterprise, as described by Minat in [MIN 99]. It aims to introduce decision-making strategies in different contexts related to the Internet as well as the types of activities to be covered in a company.
On the basis of decision-making strategies, a physical structure can be put in place and an information system architecture can be designed and implemented. In an organization, a population or a company, several architectures can be identified and implemented in their related decision-making systems. We first make a distinction between:
As a result, structures are emerging in the company’s information and decision-making system. However, it is important to highlight the presence of certain factors or contexts that will influence the entrepreneur and decision-makers in the choice of the organization to be set up. Currently, in our economy, there are some major trends that we can list [LAU 01].
They mainly concern the company’s field of action as well as the way in which business is conducted:
When we talk about a company, it is obviously a general term that refers to an industrial company as well as a bank, an insurance company or a service industry. On another level, the developing political context and global economic culture, network and transport technologies, as well as the current possibilities of information systems that allow large volumes of data to be processed at a very low cost, are leading companies to develop global strategies with greater control and flexibility.
However, this globalization that is spoken about so much is not a new fact: it has always existed, since ancient times, passing through the Greeks and Romans. Marx and Engels were already talking about it in 1848. Yet, companies are now involved in possible economic crises between Western countries (such as the USA) and Eastern countries (such as China).
Things are of course changing, especially in terms of form: the players and the technologies are different and continuously evolve; practices have been modernized and benefit from these new changes and conditions. This notion of “global (agri-)culture”, the effects of which are growing every day, is very important insofar as expectations and tastes, consumer objects and tools, technical and trading standards (e.g. through the World Trade Organization (WTO)) lead to standardizing business, products and services, as well as to strengthening capital based on the notion of profit and towards the short term, with the main customer as the shareholder.
The impact on corporate networks is significant. Thus, any innovation, in order to succeed, must be able to reach as broad a customer base as possible in order to reduce the impact of increasingly high costs and investments. The new product, or service, will therefore be designed and developed with a global perspective. In IBM’s traditional terms, we are part of a global village where markets, production and operating activities, labor and economies of scale are global. However, for the reader familiar with Kondratieff’s cycles, the globalization we are talking about today has already changed:
However, despite these trying and contradictory effects, one form of international balance or another will prevail and we will always be confronted with models of “networked distributed systems”. This requires the implementation of physical and informational communications, interactions, feedback loops, etc. And this is how the natural evolution of companies tends in turn towards more complexity! If they are subject to the complexification process that we have already described, however, like any living organism, they are subject to significant structuring effects.
In the following, we will review some structures and organizations of systems or communities of agents; we will study their characteristics and then discuss the notion of a hierarchy of levels. We will see why and how to implement these concepts in our networked companies. We will also study, among other things, an organization called “The Fractal Factory” [WAR 93] that will enable us to apply the OKP (One-of-a-Kind Production) systems principles, as experienced within IBM EMEA production systems, both at French (Montpellier) and German (Sindelfingen) manufacturing plants. Here, the difficulty comes from the economic balance between the size and the cost of each batch. Today, according to the specific applications to be covered, we would call for technologies based on, for example, 3D printing and robotics.
In any networked system, interaction is a fundamental element of the complexity that will result. Until now, the role and function of an agent has been defined, as well as its behavior. We have also talked about the types of relationships that agents could have with each other, as well as their communication protocols, but we have not yet addressed the problems of architecture, organization and structure of these relationships – what we are doing now.
It is generally agreed that a complex system is made up of autonomous agents, which commonly means decentralized and independent entities. But this is not always true: are complex systems decentralized and if so, to what extent? In terms of communications, is it the architecture that best lends itself to heterarchical interactions or does it respond to the n-cube system? These are all questions that we promise to address later on.
In Nature, to cope with complexity, the number of information and links that can be processed effectively is limited for a better control of the whole system. This problem has been solved biologically by the “multiplication of organizational levels” and the speciation of organs, as well as by a tree (and therefore hierarchical) structure that characterizes any system or network. Similarly, the corresponding control systems can be classified according to their structure. Various research studies have analyzed the evolution of the different existing structures, their advantages and disadvantages. New architectures have been proposed to improve the performance of existing industrial applications and meet the needs of future production systems. An approach that has been more widely used recently in every system since we know that most of organs or agents possess their own autonomy and by the fact that they are strongly interconnected together.
Some authors have presented the results of centralized and hierarchical controller architectures using dynamic and fully distributed or heterarchical scheduling with intelligent components. Others have proposed a classification based on four production management paradigms: centralized information (centralized decision support), distributed information (centralized decision support), centralized information (distributed decision support) and distributed information (distributed decision support).
Dilts provided an overview of the evolution of existing steering structures, from the centralized hierarchical structure to heterarchical control. He highlighted the characteristics, advantages and disadvantages of each structure [DIL 91]. He also stressed the influence and importance of a system’s architecture for the flexibility of its management and control.
Structures can be hierarchical, heterarchical, modular, holonic and agent-based. Three types of associated architecture can be distinguished: hierarchical, heterarchical and hybrid. In fact, in any case, a hybrid architecture based on the holonic concept seems to be a good solution to these different problems [KIM 02]. Finally, some French authors have provided an overview of the main possible architectures for the management of production systems and have distinguished the centralized, hierarchical, coordinated, distributed, decentralized and supervised distributed structures [PUJ 02].
In summary, a classification may be possible for steering structures. It involves organizations such as centralized or non-centralized, hierarchical or non-hierarchical, etc. First, steering structures can be classified into being centralized and non-centralized. Non-centralized structures include hierarchical, heterarchical and hybrid structures such as the n-cube. The hierarchical architecture splits into “hierarchical” and “modified hierarchical” structures. Heterarchical architecture can be decentralized or distributed. The hybrid architecture includes both hierarchical and heterarchical structures at the same time.
In the following, we will mention some work using these different architectures, while pointing out the advantages and disadvantages of each of them. This classification is also well-fitted to the recent economic situation generated by the technological evolution of very large companies such as GAFAM (Google, Apple, Facebook, Amazon, Microsoft) or BATX (Baidu, Alibaba, Tencent, Xiaomi).
The proposed centralized structure includes a control unit, or entity, that controls all production machines and has decision-making authority. It maintains the global information of all the entities’ activities in the system. This unit manages production, processes events in real time and synchronizes and coordinates all tasks (see Figure 7.1).
The benefits of this architecture include:
However, we can identify several disadvantages:
To overcome the disadvantages of centralized architecture, researchers have developed the concept of “non-centralization of decision” which intervenes through several types of architectures where decision control can be hierarchical, heterarchical (or decentralized) or hybrid.
The natural presence of hierarchy in a company and the structures of complex systems have led researchers to design hierarchical architectures. This structure defines a master–slave relationship between the upper and lower levels of management. Each level coordinates the control units from the lower level to the lowest level (see Figure 7.2). Each level has relationships that depend on the higher level, and domination on the lower level. Decisions are made by the central control unit.
Much work has contributed to the development and changes in the original reporting structure. A hierarchical control model for automated manufacturing systems has been defined [JAC 97]; the objective is to limit the size, complexity and functionality of individual control modules in hierarchical structures. The model works with the following five CIM (Computer-Aided Manufacturing) layers: facility, shop, cell, workstation and equipment. Each module breaks down the input command from the supervisor into simple subtasks, assigns them to the appropriate subordinate modules, manages their execution and finally provides the feedback status to the supervisor. This supervisor has several subordinates, and no direct communication between modules of the same level exists.
Within this framework, Chryssolouris et al. in accordance with standardized CIM architectures, described the MADEMA (Manufacturing Decision-Making) model, which has four levels of hierarchy: factory, job shop, work center and resource [CHR 88]. The first level represents the entire plant and controls the entry capacity of requests into the plant. The job shop level includes the work centers and assigns the work to these different groups. A work center level represents the grouping of production resources. The last level refers to production resource units. MADEMA receives manufacturing requests (type, quantity, due data, etc.) from the workshop level, determines the possible alternatives of the resource task pairs, the appropriate criteria, their consequences with multiple criteria, the decision support rules and finally chooses the best alternative.
Compared to operation research approaches, the MADEMA model allows for better practical and comprehensive implementations in industry. However, both models lack responsiveness and good real-time performance in the face of unforeseen events. This model was used in the early 1980s in IBM Europe’s factory management systems [MAS 89].
More recent hierarchical structures can be represented by new structures called “modified hierarchical” models. They are mainly involved by an improved control system. They enable communication and coordination between entities at the same hierarchical level. Examples of this category include “Manufacturing Systems Integration” (MSI) [SEN 94], “Production Activity Control” [AND 97] and “Factory Activity Control” (FACT) [ARE 95].
At the end of this review, we can identify the following advantages and disadvantages.
The hierarchical structure was adopted almost systematically in large systems until the 1980s. The main advantages of this structure can be summarized as follows:
Most hierarchical architectures require a fixed structure during system operation and assume the deterministic behavior of the components. These rigidities generate the main disadvantages of hierarchical architectures, which can be summarized as follows:
For this category of structure, it might be interesting to first note that the noun “heterarchy” and its adjective “heterarchical” are actually neologisms. The term heterarchical is formed from two Greek roots: heteros (other) and arckhein (to command), which originally meant “command by others”.
A heterarchy refers to the idea of different actors who assume in collegiality the coordination of a given collective action and are essentially opposed to the term hierarchy [TRE 02]. The heterarchical structure is also called the decentralized structure. In this structure, there is no higher-level control unit to coordinate all units (see Figure 7.3).
Since the control units are multiple and interacting, they can self-organize to ensure overall consistency in tasks. These units have the following four properties [CHO 93]:
From this architecture, an “egalitarian” system structure, called “peer-to-peer”, can be derived. Each element, or agent, participates in the decision-making process and enables orders to emerge. One difficulty concerns the management of restraints, deadlocks, as well as the expression of dominant choices.
In the field of network architectures, the structure described above corresponds to the so-called “peer-to-peer” connection mode. In such an organization, agents exchange and process information on the principle of equality: everyone is equal. However, this organization has two disadvantages:
Nevertheless, there is an interesting compromise, a theory that we will not describe here; it is the n-cube structure. If we refer again to the connectivity of any graph as K, the number of vertices (or agents) that can be considered is: N = 2K. This type of network offers the greatest reliable access for a given neighborhood. Indeed, if we consider that the best compromise (in terms of number of attractors and cycle length) is obtained with low connectivity, we can then consider that it is the neighborhoods of Hopfield & Moore that are most suitable for self-organization phenomena.
This last organization is widely used in information systems in order to provide a well-balanced communication system that is able to ensure an efficient and sustainable architecture.
In any production and decision system, issues of efficiency require us to continuously evolve. Indeed, the governance of any system is submitted to the changes in the global environment. Thus, the architectures and structures described above may satisfy most of the common requirements arising from business, social, economic or customer needs.
However, it is of the highest importance to adapt these architectures [MAS 18] with the new challenges raised in our economy, such as: the meta-governance principles applied to the management of the economy by a few large countries, and then the upper-governance over imposed by large and monopolistic companies such as GAFAMs and BATXs.