In this paper, a framework of distributed fault detection for network systems is developed. Specifically, a framework of distributed cyber attack detection system for synchronized large-scale power network is constructed, where active power flow in power network system is modeled by the swing equation and cyber attacks are modeled as unknown power generation or consumption. The approach is based on fault detection and identification (FDI) filter so that malicious cyber attacks in the neighborhood of a node are identified through local information that consists of local power consumption, generation, and power flow. The FDI filter is a special Luenberger observer whose parameter is selected in such a way that residual between the sensed power flow and the FDI filter's output is only affected by specific cyber attacks. Residual indicates the existence of the cyber attack. A sufficient condition is provided for the existence of the FDI filter with the local input and the output to detect fault in the network system. A numerical example is provided to demonstrate efficacy of the proposed approach.