A generic sensor value validation procedure for knowledge based fault diagnosis systems

The fault diagnostic task is a critical aspect of the operation of sophisticated physical systems. Knowledge based systems are increasingly being em ployed for monitoring and fault diagnosis, in view of the complexity of these systems. In most cases, sensors are the only means of observing the operating state of the physical systems for both human operators and knowledge based systems. Validation of these sensor values is an important prerequisite for safe and reliable monitoring and diagnostics functions. In this thesis, a generic sensor value validation procedure for knowledge based fault diagnosis systems is enhanced and evaluated. The validation procedure is based on systematically exploring all the embedded sensor redundancies in a physical system to a maximum extent. The concept of causal relations and sensor redundancy graphs are described. Validity of sensor values are represented by defining validity levels. The procedure for validating sensor values with different precision levels is discussed. The validation of binary sensors is also described. A conceptual frame work for interfacing a validation procedure with knowledge based systems is discussed. Two physical systems are described and results of the implementation of the procedure as applicable to them are shown.