A methodology for incipient fault detection

This paper proposes a fault detection methodology for incipient faults that combines different residual generation methods (observers and l-step ahead predictors) with different convergence velocity to the real output trying to benefit from the advantages offered by each one. The integration is based on generating a timed automaton, which combines the information extracted from each method in order to provide the best fault detection performance regarding incipient faults. The proposed methodology has as a main objective to detect as early as possible anomalies or incipient faults in system components. Nowadays, for many systems, early warnings contribute to increase system reliability, prevent major component failures and planning the necessary repair actions for several weeks (predictive maintenance). The application of this methodology will be illustrated in a case study based on a part of the Barcelona water network.