Fast-convergent fault detection and isolation in a class of nonlinear uncertain systems

Abstract

The present work proposes a fast-convergent fault detection and isolation (FDI) scheme for linear systems affected by model uncertainties, such as unknown inputs or unbounded nonlinearities. The finite-time convergence is attained by transforming the I/O signals through Volterra operators with suitably designed kernel functions. A novel feature of the proposed approach is the exploitation of a system decomposition that allows removing the effect of intractable uncertainties while recasting the system dynamics in a form applicable for Volterra operators to achieve non-asymptotic estimation. Remarkably, the proposed approach can reconstruct the state variables of the system in an arbitrarily short time and the fault can be diagnosed efficiently by imposing detection and isolation thresholds on transformed signals. The detectability and isolability of the fault are also characterized. The proposed FDI scheme is applied in simulation to a web process system to diagnose the presence of actuator faults. Simulation results confirm the effectiveness of the proposed scheme in two scenarios with nonlinear uncertainties. Furthermore, comparisons are made between the proposed method and a Sliding Mode Control (SMC) method in terms of estimation performance and computational complexity.