Abstract: Based on the decentralized computing platform, a decentralized and self-organizing algorithm for sensor fault diagnosis is proposed. The basic idea of the method is to detect the equality and inequality constraints of the adjacent variables under the hypothesis that the fault sensors are always less. Unlike traditional centralized and distributed algorithms, the algorithm does not need to build a central node, without increasing hardware or time-series redundancy, and regardless of spatial similarity. This method is very effective o detect the fault of Single or multiple sensors through simulation result.