Data-driven approach augmented in simulation for robust fault prognosis

The implementation of data-driven methods of fault prognosis requires the availability of data describing the degradation process that can be used for learning. These data are rarely available, and when available they do not represent all the possible operating conditions of the considered system. This paper proposes a data-driven method based on data augmented in simulation to take into account all the possible trends of the degradation process of a system, then, the data set is used for an offline estimation of the Remaining Useful Life within a confidence interval that gives to the user a security margin. The robustness of the Remaining Useful Life estimation to the changes in Condition Monitoring is performed online by updating the model parameters. The trend model used for the estimation of the Remaining Useful Life is a Wiener process whose drift parameter is updated online and whose stochastic part is used to generate in simulation a set of degradation trajectories. Simulations and experimental results obtained on different systems show the effectiveness and the wideness of the application scope of the proposed approach.