Multi-agent systems (MAS) are particularly adapted to deal with dynamic distributed environments and are typically used to manage business workflows or data flows in manufacturing systems. From a control perspective, holonic representation allows both the informational and the physical parts in such manufacturing systems to be modelled. In this article, we focus on the notion of role in holonic MAS (HoloMAS) and its contribution to the adaptive control of manufacturing systems. We validated our HoloMAS proposal using simulations and through a real implementation on a flexible assembly cell in our university lab. Our results clearly show that the notion of role helps to adapt and manage perturbations.