Synthesis of quasi-redundant sensor data: a probabilistic approach

It is common in industrial application that multiple sensors are placed to measure similar parameters. These sensors may not be truly redundant in the sense that they measure exactly the same parameter, but the phenomena they do measure may be very well correlated. We denote such sensors as being quasi-redundant. This papers concerns a methodology for fusion of such quasi-redundant sensors. The goal of this new methodology is to produce a single inferred, high-confidence measure from data obtained from multiple such sensors. The techniques developed in this paper exploit the quasi-redundant relationship between sensors (1) to replace missing or low-confidence data and (2) to synthesize complementary sensor data to improve model development and testing. These techniques are illustrated using measured blast temperature data from a cupola iron-melting furnace.