Sensor fault diagnosis with a probabilistic decision process

In this paper a probabilistic approach to sensor fault diagnosis is presented. The proposed method is applicable to systems whose dynamic can be approximated with only few active states, especially in process control where we usually have a relatively slow dynamics. Unlike most existing probabilistic approaches to fault diagnosis, which are based on Bayesian Belief Networks, in this approach the probabilistic model is directly extracted from a parity equation. The relevant parity equation can be found using a model of the system or through principal component analysis of data measured from the system. In addition, a sensor detectability index is introduced that specifies the level of detectability of sensor faults in a set of analytically redundant sensors. This index depends only on the internal relationships of the variables of the system and noise level. The method is tested on a model of the Tennessee Eastman process and the result shows a fast and reliable prediction of fault in the detectable sensors.

► Data driven probabilistic approach to sensor fault diagnosis.
► Sensor detectability index (SDI) based on the internal relationships of the system variables.
► Application to Tennessee Eastman process to validate approach.
► Gaussian model found effective in representing noise.
► Approach extendable to nonlinear systems using Mixture of Probabilistic Models.