Adaptive autoregressive modeling of non-stationary vibration signals under distinct gear states. Part 1: modeling

Non-parametric time–frequency techniques are increasingly developed and employed to process non-stationary vibration signals of rotating machinery in a great deal of condition monitoring literature. However, their capacity to reveal power variations in the time–frequency space as precisely as possible becomes a hard constraint when the aim is that of monitoring the occurrence of mechanical faults. Therefore, for an early diagnosis, it is imperative to utilize methods with high temporal resolution, aiming at detecting spectral variations occurring in a very short time. This paper proposes three new adaptive parametric models transformed from time-varying vector-autoregressive model with their parameters estimated by means of noise-adaptive Kalman filter, extended Kalman filter and modified extended Kalman filter, respectively, on the basis of different assumptions. The performance analysis of the proposed adaptive parametric models is demonstrated using numerically generated non-stationary test signals. The results suggest that the proposed models possess appealing advantages in processing non-stationary signals and thus are able to provide reliable time–frequency domain information for condition monitoring.