Reduction of detectable fault size in sampled data residuals

One of the families of robust algorithms for fault detection in dynamic systems is related to the use of mathematical models. This poses challenges when access to the mathematical model for design is not available. This work employs a scheme capable of diagnosing systems based on input-output data (designed under nominal operation and the assumption that the system exhibits essentially linear behavior). A relevant aspect of diagnosis is the sensitivity to the faults being detected. This sensitivity plays a crucial role in critical systems, where early fault detection is desired. This work offers an approach capable of adjusting the diagnostic sensitivity of the residual-generating algorithms with respect to faults. The idea is to obtain residuals that are triggered once a certain predetermined threshold value is exceeded due to the effect of faults. Based on an application example, it is illustrated that the proposed residual generator is approximately 20 times more sensitive than those reported in the literature.