Deterministic particle filtering and application to diagnosis of a roller bearing

In this article, the detection of a fault on the inner race of a roller bearing is presented as a problem of optimal estimation of a hidden fault, via measures delivered by a vibration sensor. First, we propose a linear model for the transmission of a vibratory signal to the sensor's diaphragm. The impact of shocks due to the default is represented by a stochastic drift term whose values are in a discrete set. To determine the state of the roller bearing, we estimate the value of this term using particular filtering.