A real-time sequential filtering fusion algorithm and a fault diagnosis method are proposed for the networked multi-sensor systems with communication constraints. Firstly, the local optimal filter is obtained for the sensors which have access to the filter center. The Kalman predictor is used for the sensors left to obtain the local optimal prediction according to the latest measurement. The optimal fusion estimation is derived from the fusion criterion weighted by matrices. The corresponding fault diagnosis method is used to locate the fault sensors. Finally, a target tracking example is given to illustrate the effectiveness of the proposed method.