A Layered, Any Time Approach to Sensor Validation

Sensors are the most usual source of information in many automatic systems such as automatic control, diagnosis, monitoring, etc. These computerised systems utilise different models of the process being served which usually, assume the value of the variables as a correct reading from the sensors. Unfortunately, sensors are prone to failures. This article proposes a layered approach to the use of sensor information where the lowest layer validates sensors and provides the information to the higher layers that model the process. The proposed mechanism utilises belief networks as the framework for failure detection, and uses a property based on the Markov blanket to isolate the faulty sensors from the apparently faulty sensors. Additionally, an any time version of the sensor validation algorithm is presented and the approach is tested on the validation of temperature sensors in a gas turbine of a power plant.