This paper addresses the problem of data fusion in decentralized senso
r networks in which there is no central fusion center and in which no
single sensor node has global knowledge of network topology. An algori
thm is described which is able to locally process and assimilate data
in such networks and which yields an identical result to that obtained
in a centralized system. An implementation of this decentralized data
fusion algorithm on a process control rig incorporating over 150 sens
ors and 30 processors is described. The results show that the decentra
lized data fusion system described in this paper offers many advantage
s in terms of robustness, scalability and flexibility over a centraliz
ed system.