Fault detection, isolation, and recovery using spline tools and differential flatness with application to a magnetic lévitation system

This paper discusses fault detection and isolation for continuous-time systems using B-Splines and the notion of differential flatness. The idea is, from the system's flat outputs (which are obtained directly from measurement or from an observer), we algebraically produce every other measured signal, including the inputs. The corresponding signals are then compared. In nominal condition, a measured signal and its counterpart derived from the flat outputs are similar up to noise and the filter's bandwidth. In the occurrence of faults, they are different. We then use this information to signify a fault and to compensate for it. The techniques used to produce signals from the flat outputs, and filter out the noise, are based on a B-Splines parametrisation.