A discrete-time sliding mode observer for estimation of auto-regressive model coefficients with an application in condition monitoring

Abstract

Development of a sliding mode observer (SMO) scheme for on-line condition monitoring of dry vacuum pumps is considered. The exhaust pressure signal from such a pump can be practically acquired with a standard transducer, and described with an auto-regressive (AR) model. A novel discrete-time SMO scheme has been designed to estimate AR model coefficients based on a short data set sampled from the exhaust pressure signal, and a nominal set of model coefficients estimated from fault-free data. Vacuum pumps' exhausts are at risk of blockage due to solid deposits of process chemicals. The results demonstrate that the reduction in free volume of the silencer can be detected by monitoring the injection signal of the SMO. The magnitude of the injection signal is related to the difference in location between the poles of the nominal AR model and those of the estimated model