Finite-Time Sensors Diagnosis Based on Unknown Input Observers

The paper focuses on developing a finite-time approach for sensors fault detection and isolation in MIMO systems. The proposed solution utilizes the synthesis of a bank of unknown input observers, tailored specifically for each of sensor. A significant advantage of this solution over existing methods is its applicability to plants with any arbitrary relative degree. The paper introduces a modification that ensures finite-time convergence using Kreisselmeier's scheme combined with the dynamic regressor extension and mixing method. This modification guarantees that the convergence of observer estimates—and thus the diagnostic results—occurs within a user-specified timeframe. Additionally, the implementation of filters and threshold settings is discussed to prevent false alarms caused by measurement noise. The effectiveness of the developed approach is validated through computer simulations, confirming its robust performance.