Modern theories propose autonomous structures as building blocks of next-generation manufacturing systems. However, their size and scope are not agreed upon and remain a subject of research. The paper presents a method for discovering autonomous structures within existing manufacturing systems. Firstly, it is shown how a complex network model of a manufacturing system can be obtained. Then, a method for discovering structure in complex networks is applied in order to find cohesive subnetworks – candidates for the formation of autonomous work systems. The approach is illustrated in a case study of engineer-to-order production.
In today's turbulent competitive environment, manufacturing systems’ agility, flexibility, and adaptability are key requirements for success. However, most manufacturing companies still exhibit rigid organisational structures with highly centralised control. As an alternative, autonomous control has been suggested by numerous authors, who hypothesise that autonomy leads to better control of the system's emergent properties. Furthermore, better complexity management achieved this way leads to better system performance in highly dynamic environments [1].
The idea of autonomously controlled subsystems has existed since the 1970s, when promising results were obtained in large-scale experiments in the Swedish automotive industry. In the case of Saab, 600 workers were organised into autonomous teams, which led to an increase in productivity and quality, a decrease in machine breakdowns, a decrease in costs, and positive social effects [2].
Although success in automotive industry was short-lived because of the nature of serial production, recent manufacturing trends [3] such as decreasing lot sizes and personalisation of products have led researchers to revisit these ideas.
Modern approaches such as holonic manufacturing systems [4], autonomous work systems [5], and bionic manufacturing systems [6] all conceive of the production environment as a network of autonomous elements, working together through mechanisms of coordination and cooperation. However, the size and scope of autonomous structures are not agreed upon and remain a subject of research. Should the basic autonomous manufacturing unit be a workshop, a group of work systems working together on the same product, a manufacturing cell, or a single work system? Moreover, on what basis should autonomous units be formed? These issues are particularly relevant in one-of-a-kind and engineer-to-order (ETO) types of production where orders arrive randomly and with almost no repetitions.
The paper addresses these questions from the perspective of complex networks theory. The theory allows us to reconsider the issues of complexity and autonomy in order to find a better way of structuring a manufacturing system as a network of autonomous work subsystems. If a subsystem is to be autonomous, it must have the ability to self-organise. In a self-organising system, the information flow within the system must exceed the one between the subsystem and its surroundings. If this is to be true, elements of the subsystem must be strongly connected in comparison to their connections with external elements. On this foundation, we assume that it is possible to identify autonomous subsystems with complex network analysis.
The paper shows how a manufacturing system can be viewed as a complex network of elementary work systems, and how this network can be extracted from the manufacturing execution system (MES) data. Autonomous structures are then sought through clique percolation (CPM) [7], a social network analysis method. Identified sub-networks are used to guide an expert through the process of autonomous work system formation.
An industrial case study of ETO production is presented. The results show that existing as well as new autonomous structures are identified and suggest that the approach could be used to improve factory structure and layout as well as organisation of location-independent work such as manual welding of large workpieces or construction site assembly and installation.
The paper presents a method of discovering autonomous structures in complex networks of work systems. In essence, it shows (1) how a network representation of a manufacturing system can be acquired from MES data, (2) how the network can be proven to be complex, and (3) how autonomous structures within the network can be discovered.
Future work will focus on further generalisation of results through subsequent case studies. Further steps are also needed for practical implementation of autonomous work systems. Another aspect that needs to be addressed in the future is the changeability of system structure through time. We foresee that the presented method will be able to assist in the creation of ad-hoc work structures needed for location-independent work such as manual welding of large workpieces or construction site assembly and installation. Lastly, the paper highlights the potential of network analysis for manufacturing applications.
