Abstract-The objective of the present paper is to provide a methodology where the functional characteristics of a distributed engineering system can be obtained by merging domaindependent knowledge at run-time. We focus on distributed control systems where computing nodes are related to the physical environment in which they operate via sensors/actuators. The knowledge representation is formally expressed with a mereological approach where a structural mereology describes the physical environment and a functional mereology identifies available engineering goals for each computing node. During the design step, a mechanism based on Formal Concept Analysis (FCA) allows to generate the resulting goal mereology. The concept of goal is refined with sub-concepts in the multilevel structure. Because computing nodes depend on each other for goals to be achieved, an agent-based method is proposed to establish dynamically the dependencies among distributed nodes. This method is centered on a fusion mechanism involving the functional mereologies of appropriate nodes. We use an example from an open-channel hydraulic system controlling the water level to motivate and illustrate the model. Although it is limited to the engineering systems, this approach can be reused in related domains where the goal representation can be expressed as a triple including an action, a role and a physical entity.