© 2018 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works.

The main objective of this paper is to propose an on-line statistic algorithm for fault detection in non-linear dynamic systems based on data analysis. This discipline is a branch of the statistical science and it allows to organize information contained in a data set to make an inference about the origin of this information. Our strategy is founded in the common fault detection techniques based on system modeling. Basically, these techniques produce a residual signal, which is studied to identify when a fault occurs in a given system. In this paper, this residual signal is analyzed through a basic on-line statistic parameter: the variance. To fulfill our main objective, a healthy threshold is obtained such that if it is overcome, it can be assumed that a fault is detected. The fault may be, for instance, a change in a system's parameter. Our strategy is specifically applied to an adaptive predictive controlled base-isolated system. This controlled system is employed to mitigate vibration in buildings when an external seismic perturbation exists. Additionally, numerical experiments validate that the proposed fault detection strategy is functional when a modeled fault is inserted to the system. These numerical experiments are realized by employing four different real earthquake data and by using an artificial earthquake here developed.

Peer Reviewed