8.1.1.1. Flexible and reconfigurable workshops

Various studies have examined decentralized structures in production systems. Their contributions are in the development of the architecture, the design of negotiation protocols between entities and industrial applications. The use of heterarchical and cooperative architectures are alternatives to hierarchical architecture [HAT 85]. In this context, it has been possible to define, through IBM’s PIAUL project, “expert” rules of behavior, local goals and global goals (that autonomous entities follow) in order to prevent anarchy and chaos in the system.

Shaw described a distributed control structure for dynamic scheduling in a cellular manufacturing system (CMS) [SHA 87]. The architecture of a CMS consists of three types of units or “cells” (warehouse cell, pallet cell and robot cell) (see Figure 8.1). Each cell acts independently by exchanging messages, and each cell controller maintains this local information, but without global control. The assignment of tasks is carried out dynamically by negotiation between the cell controllers. The scheduling of tasks in each cell is done locally. Recent work by Kondoh et al. has proposed a heterarchical structure for the CMS [KON 00, TOM 97, MAS 01d]. These authors consider the CMS (Cellular Manufacturing System) principle as a rapid prototype for design and as a decision support tool for configuring and assigning tasks at the resource and product levels.

Configurations and assignments of parts and resources are determined by self-organization mechanisms between the different entities at operational level. This evolution towards more autonomy for production units naturally leads us to consider this approach as a framework for the application of heterarchical control systems.

Coordination between the independent entities of a heterarchical structure is an essential point to be considered when managing a system. We can ensure this through predefined rules (centrally or not); we can also leave this coordination “open”. This is generally the case with our approach and was implemented in network structures where agents use communication protocols based on market paradigm bidding principles to meet their objectives. The canonical example of a standard negotiation protocol is the Contract Net Protocol (CNP) developed by Smith [SMI 80]. It has been widely used by various works using heterarchical architectures, and has been extended through different auction mechanisms [FIP 00, LIU 02] and negotiation protocols.

Hence, the heterarchical architecture is used for the management of a production workshop in which products and resources are considered as agents. Each agent negotiates with the other agents, in real time, through the principle of an exchange market, in order to satisfy its individual objectives [LIN 92]. When a customer requests a service provided by an organization, a cost in exchange currency is required by the organization. This model uses a generic construction mechanism for exchange offers during negotiation between agents, based on the principle of combining price and objective (time, cost, quality, etc.). This architecture, through simulations, reveals a great flexibility and adaptability in the real-time management of a production workshop.

A decentralized architecture for the management of a production workshop [PAR 98a] with resources, a manager, product type and processing unit has been developed with intelligent entities or intelligent agents who know “how to combine products and resources for the manufacture of other products”. In this structure, the authors provide a mechanism for direct dialogue between clients and the production workshop for “mass customization”, using intelligent agent technology.

Finally, the European PABADIS project – Plant Automation Based on Distributed Systems – was a recent example of a heterarchical architecture designed for industrial applications. The PABADIS system uses a decentralized organization for the automatic and dynamic reconfiguration of production lines [PAB 00]. It aims to improve the management of a decentralized production system by using the notion of “plug-and-participate” and the total or partial elimination of planning and scheduling tasks. The basic components in PABADIS are agents and services. Agents and services cooperate to accomplish the tasks to be performed.

8.1.1.2. Hybrid steering structures

The hierarchical and heterarchical structures we have described have both advantages and disadvantages for the management of distributed production systems. For this reason, some research has tried to preserve the advantages of both structures by proposing a new “hybrid” structure (Figure 8.2).

In the hybrid structure, control units of the same hierarchical level are interconnected via the same control fixture. They are able to communicate and cooperate to meet their local objectives. During disruptions, all control units can ask their checking fixtures for help in solving the problems detected. In this model, control units require the assistance of higher-level control units in the event of non-compliance with their local objectives due to unexpected disruptions. Ottaway and Burns propose an adaptive production control system (APCS), similar to the so-called CAS systems, as seen in Chapter 6. Within this adaptive context, the transition between heterarchical and hierarchical structure occurs dynamically and is based on the system’s workload [OTT 00]. In their models, these authors consider the three types of agents: task, resource and supervisor agents. Each agent has knowledge of the coordination, production, interface and inference engine. When an agent notes that the resource it represents is not being properly used, it asks for a monitoring agent to control the resource. In this way, a hierarchy level is dynamically introduced into the system.

MetaMorph, a hybrid agent-based architecture for controlling distributed production systems [MAT 99], uses two types of agents, the resource agent to represent physical resources and the mediator agent for coordination between resource agents. We have added hierarchical mediators and cooperative negotiation mechanisms between resource agents to our model. Recent hybrid structures include the holonic manufacturing system (HMS). The basic principles of holonic systems were introduced in 1967 by Arthur Koestler in his book The Ghost in the Machine [KOE 67]. Koestler introduced the idea that a few key principles were sufficient to explain the ability of social and biological systems to regulate themselves. He proposed the term “holon” to describe the basic element of these systems. This word combines the Greek root “holos”, which means “whole”, with the suffix “on”, or “part”, as for proton or neutron in an atom.

The HMS system appears more like the search for a compromise between integrated or hierarchical organizations, on the one hand, as already coherent units within a system, and on the other hand, distributed or heterarchical organizations as reactive parts to the environment of this system. This interpretation is recent [VAL 99] for the concept’s production systems. Its purpose is to satisfy the adaptability criteria required by the new generation of production systems (Next Generation Manufacturing System) [KUR 96]. It was generally introduced by the IMS (Intelligent Manufacturing Systems) initiative launched by Japan in 1989 [KIM 97]. The “holarchical” structure resulting from the arrangement of these production holons can be considered as the compromise between a hierarchical structure and a heterarchical structure insofar as the cooperation of low-level “intelligent” holons remains nevertheless coordinated by hierarchically superior holons.

Brussel et al., Valckenaers et al. and Wyns present the PROSA architecture (Product-Resource-Order-Staff Architecture) as a holonic structure for production management [WYN 99]. Among the examples of holonic architecture is the PROSA (Product-Resource-Order-Staff Architecture) for production management, developed by the research group at the Université Catholique de Louvain in Belgium [BRU 98]. PROSA is a holistic HMS production system designed to achieve stability despite disruptions, flexibility, adaptability to change and efficient use of resources. This architecture includes three basic types of holons: order, product and resource, as well as a staff holon. Each of the basic holons is responsible for logistics, technological planning (including process planning) and the determination of the resource’s capabilities, respectively. The staff holon is considered an external expert who provides advice to the basic holons. They can provide centralized algorithms for scheduling and help basic holons. By including staff holons in the structure, the system has a hierarchical management behavior that can improve its overall performance.

8.1.1.3. Discussion

After examining the characteristics of current research on organizations, as well as the advantages and disadvantages of the various existing steering structures, we find that there is no ideal generic steering model that can be used at any time and in all environments. Each structure can be effective for certain types of problems and environments depending on the context, dynamic and time constraints, etc. We should also note that, contrary to what specialists frequently think, the physical structure of a system can be organized in one way and its information system in another. Thus, a hybrid network or an n-cube system does not exempt from prioritizing information, otherwise the system will quickly overload. In this case, the frequent removal of hierarchical levels is not necessarily the right approach to reduce complexity.

Our work here focuses on self-organization mechanisms between autonomous entities through negotiation protocols and cooperation mechanisms (e.g. remember that coopetition = cooperation + then competition). These approaches can be applied for the dynamic allocation of resources in a dynamic and situated environment, between the product and resource entities of a production workshop and considered locally. The most appropriate architecture for these concepts is a heterarchical one. Its specific characteristics have allowed us to apply and validate different concepts and mechanisms with more freedom of interaction and simplicity of modeling. These concepts have been developed to involve interactions between autonomous entities in located dynamic environments. However, we can also apply these concepts with hybrid control architectures and with a high degree of autonomy of the production units. As a practical example of heterarchical architecture, we mentioned the European PABADIS project, in which our LGI2P-EMA research center at the Ecole des Mines in France was a partner. This architecture allowed us to test and validate the concepts developed.

In the following sections, we will explain the techniques and algorithms that can be used to control heterarchical production systems.

Discussion

The configuration, or dynamic reconfiguration, of distributed production systems should take into account the criteria set out above. However, there is one condition that has not yet been met and it is the one that leads us to self-organization. In a simplified way, self-organization is a process of selecting and eliminating the worst performers. This selection is accompanied by local variations and optimization, which, in the scheme of evolution and natural selection, only explains minor adaptations and not qualitative leaps. Cooperation and immediate need are not sufficient conditions and in fact require competition between groups, with adaptation strategies, in order to have the best access to resources.

This therefore requires, as we saw in a previous chapter on the emergence of chaos, dissipative structures, i.e. structures capable of diffusion: a phenomenon that makes it possible to homogenize components, inhibit or activate characters, in short to create differentiation, forms and order. This order therefore emerges from the interactivity of the entities and does not pre-exist to the entities that constitute it.

What forms a society and structures the relationships between its entities is therefore social rules, or meta-rules, which define a framework for action, spaces of freedom or instructions. These rules also consist of methods or instructions for use that allow the system to operate and evolve in a reverse mode.

Thus, in fractal systems or architectures, and because of their invariance properties, there are tree structures that limit the system complexity. This is above all a characteristic of finalized systems much more than self-organization. This feature allows great flexibility and adaptation, but it is not enough. It is therefore necessary to add autonomy or self-management at each level.

Similarly, a dissipative structure cannot be hierarchical. However, in the case of the fractals of interest to us, they are not self-organized but result from an external force with a long range (general organizational laws) that meets local resistance (context and local constraints). This fragments and organizes them in such a way that organized autonomy then occurs.