Self-organisation of mobile robots in large structure assembly using multi-agent systems

Competition between manufacturers in large structure assembly (LSA) is driven by the need to improve the adaptability and versatility of their manufacturing systems. The lack of these qualities in the currently used systems is caused by the dedicated nature of their fixtures and jigs. This has led to their underutilisation and costly changeover procedures. In addition to that, modern automation systems tend to be dedicated to very specific tasks. This means that such systems are highly specialised and can reach obsolescence once there is a substantial change in production requirements. In this doctoral thesis, a dynamic system consisting of mobile robots is proposed to overcome those limitations. As a first knowledge contribution in this doctoral thesis, it is investigated under which conditions using mobile robots instead of the traditional, fixed automation systems in LSA can be advantageous. In this context, dynamic systems are expected to be more versatile and adaptive than fixed systems. Unlike traditional, dedicated automation systems, they are not constrained to gantry rails or fixed to the floor. This results in an expanded working envelope and consequently the ability to reach more workstations. Furthermore, if a product is large enough, the manufacturer can choose how many mobile robots to deploy around it. Accordingly, it was shown that the ability to balance work rates on products and consequently meet their due times is improved. For the second knowledge contribution, two fundamentally different decision-making models for controlling mobile agents in the complex scheduling problem are investigated. This is done to investigate ways of taking full advantage from the potential benefits of applying mobile robots. It is found that existing models from related academic literature are not suited for the given problem. Therefore, two new models had to be proposed for this purpose. It was plausible to use an agent-based approach for self-organisation. This is because similarly to agents, mobile robots can perform independently of one-another; and have limited perception and communication abilities. Finally, through a comparison study, scenarios are identified where either model is better to use. In agreement with much of the established literature in the field, the models are shown to exhibit the common advantages and disadvantages of their respective architecture types. Considering that the enabling technologies are nearing sufficient maturity for deploying mobile robots in LSA, it is concluded that this approach can have several advantages. Firstly, the granularity and freedom of movement enables much more control over product completion times. Secondly, the increased working envelope enables higher utilisation of manufacturing resources. In the context of LSA, this is a considerable challenge because products take a very long time to get loaded and unloaded from workstations. However, if the product flow is steady, there are rare disruptions and rare production changes, fixed automation systems have an advantage due to requiring much less time (if any) for moving and localising. Therefore, mobile systems become more preferred to fixed systems in environments where there is an increasing frequency of disruptions and changes in production requirements. The validation of agent-based self-organisation models for mobile robots in LSA confirms the expectations based on existing literature. Also, it reveals that with relatively low amounts of spare capacity (5%) in the manufacturing systems, there is little need for sophisticated models. The value of optimised models becomes apparent when spare capacity approaches 0% (or even negative values) and there is less room for inefficiencies in scheduling.