Condition monitoring of rotating machinery using cyclic autoregressive models

So far the modeling of vibrations for fault diagnosis in machinery has been achieved using time invariant autoregressive models which assume the signal to he stationary. The periodic nature of rotating machinery however gives rise to vibrations which are cyclostationary. This paper applies a periodically time varying autoregressive fitter to better model such signals. Experimental results indicate that a model based approach can be applied to fault diagnosis where vibration data from faulty machines are available and to fault detection where only vibration data from normal operation are available.