Ubiquitous monitoring over wireless sensor networks (WSNs) is of increasing interest in industrial cyber-physical systems (CPSs). Question of how to understand a situation of physical system by estimating process parameters is largely unexplored. This paper is concerned with the distributed estimation problem for industrial automation over relay-assisted WSNs. Different from most existing works on WSN with homogeneous sensor nodes, the network considered in this paper consists of two types of nodes, i.e., sensing nodes (SNs), which is capable of sensing and computing, and relay nodes (RNs), which is only capable of simple data aggregation. We first adopt a Kalman filtering (KF) approach to estimate the unknown physical parameters. In order to facilitate the decentralized implementation of the KF algorithm in relay-assisted WSNs, a tree-based broadcasting strategy is provided for distributed sensor fusion. With the fused information, the consensus-based estimation algorithms are proposed for SNs and RNs, respectively. The proposed method is applied to estimate the slab temperature distribution in a hot rolling process monitoring system, which is a typical industrial CPS. It is demonstrated that the introduction of RNs improves temperature estimation efficiency and accuracy compared with the homogeneous WSN with SNs only.