Diagnostic models for rotating machinery subject to vibration monitoring for condition-based maintenance

Vibration monitoring by measuring at a remote station some vibration signals presents a unique and appealing means to conduct condition monitoring for rotating machinery. Nowadays, the time-varying spectrum of non-stationary vibration signals defined in the time-frequency domain has increasingly impelled the development of time-frequency techniques in the state diagnosis of rotating machinery. However, when the capacity to reveal power variations in the time-frequency space as precisely as possible and independence of varying load conditions are incorporated, to develop such a robust diagnostic technique for on-line condition monitoring purposes becomes more complicated.; This thesis attempts to solve the above problems by means of adaptive parametric modeling. It focuses on the adaptive parametric modeling of non-stationary vibration signals and performance analysis in the time-frequency domain under distinct states of gearbox, the improvement in time-frequency representation, and the development of load-independent state indices for on-line maintenance decision model. The techniques of time-varying autoregressive models by means of advanced Kalman filtering algorithms under distinct assumptions are proposed and show remarkable advantages over conventional non-parametric time-frequency techniques. Two robust state indices, termed bispectral-domain feature energy and autoregressive model residual-based machine state deterioration parameter, are developed and demonstrate remarkable advantages over other newly-proposed state indices of rotating machinery primarily with respect to the enhancement of load-independence and fault-induced effects. These two state indices can be directly employed by on-line maintenance decision model.; In this thesis, both simulated signals and full lifetime inspection data of gearbox from a variety of test-runs are used and we will attempt to provide complete analysis of the sensitivity and robustness of the techniques proposed herein.