System identification for multi-sensor data fusion
Karla Hernández, James C. Spall · 2015
In this paper we discuss the problem of combining sensor information for two main detection problems: 1) two variants of a spatial search problem and 2) a fault detection problem for a three tank system (TTS). In all cases the assumption is that data may be collected from multiple sensors. The goal is then to combine all the information to determine whether an object (or fault) is present in a given area. In our setting there exist two main types of sensors, namely: “small” and “large” sensors. Essentially, it is assumed that small sensors can inspect an area that is relatively small in comparison to that which the large sensor can inspect. By deriving a relationship between small and large sensor measurements we combine data using a maximum likelihood based methodology. In particular, each detection problem is initially formulated as a system identification problem. Here, the large sensor collects data on the full system while small sensors collect data on subsystems. By establishing a connection of this identification problem to existing literature, we can obtain asymptotic convergence and asymptotic normality results.