The condition monitoring system based on sensor network becomes widely used among various fields. The sensor data is utilized to monitor the present condition, to predict future condition and to detect the state change point from the behavior of sensor data. The behavior of sensor data before reaching the abnormal state would interact with other behaviors of sensor data if the sensors monitor the same object. In this paper we focus on the relationship between sensor data represented by the graphs. The sequences of the relationship from the change detection time are stored in the database. We propose that the state change detection is determined by searching the similar sequence of the relationship.