| This work formulates the problem of incipient fault detection and diagnostics for rotating machinery in a statistical model-based framework. This includes problem description, modeling of rotating machinery and fault mechanisms, formulation of the detection and diagnostics problem, and an implementation of the proposed scheme in a simulation environment to test the feasibility of this approach. More specifically, a multiple model nonlinear filtering algorithm is proposed for fault detection and diagnostics in a statistical framework. The development of stochastic nonlinear observers, which are used as fault filters, is presented. A constant observer gain, which guarantees asymptotic performance, is derived for the nonlinear stochastic observers using concepts of estimation theory and Lyapunov stability theory. Exponential stability of the mean of the estimation error and uniform boundedness of the error variance of the nonlinear stochastic observers is proved. The stochastic observers filter the measurements (vibration signals) based on a mathematical model of the rotating machine, thereby attenuating the noise without losing the significant information in the signal. By statistically correlating the output of the filters with the actual vibration signals from the machine, it is possible to detect and diagnose faults in the rotating machinery. The implementation of the proposed fault detection scheme has been done taking into consideration different fault scenarios. A simulation study, which includes normal and different fault modes, illustrates the potential of the proposed approach, especially in the presence of measurement noise and process uncertainty. |
