Parameter estimation methods for fault detection and isolation

Fault detection via parameter estimation relies in the principle that possible faults in the monitored system can be associated with specific parameters and states of the mathematical model of the system given in the form of an input-output relation: y(t)=f(u,e,θ,x) where y(t) represents the output vector of the system, u(t) the input vector, x(t) the state variables which are partially measurable, θ the non measurable parameters which are likely to change on the occurrence of a fault, and e(t) the modeling errors and/or noise terms affecting the process. The general procedure to detect faults follows the steps below: (1) Establishment of the mathematical model of the system’s normal behavior, y(t)=f(u(t),θ) At this stage, allowable tolerances for the system’s parameter values are also defined. (2) Determination of the relationship between the model parameters θi and the physical system parameters pj. (3) Estimation of the model parameters θi from measurements of y(t), u(t) by a suitable estimation procedure,