In this paper, we propose a robustness data collection and fault detection method for large-scale sensor networks, which consist of ten thousand sensor nodes. The sensor network has some special features as multihop communication, data aggregation, and high error data rate caused by the huge number of nodes. To tackle this issue, we propose a distributed and self-organized method for data collecting and diagnosis. Our method has two features that are token nodes and limited broadcast for the data sharing and collecting. The token node is a special node that can make a limited broadcast packet for data collecting and sharing. Limited broadcast is a limited multihop number and we apply only two hops on this research. Those features can prevent a flood of broadcast packet. To clear the capability of our method, we make a few simulations and compare with other data collecting and diagnose method. As the result, our proposed method has better performance than other method.