Optimally weighted recursive median filters for denoising vibration signals used in health monitoring of rotating systems

Abstract

A novel weighted recursive median (WRM) filter for denoising vibration signals is presented in this paper. Vibration signals emanating from rotating systems are typically periodic as the transients die out with time due to damping. The weights of this nonlinear filter are optimized for typical vibration signals encountered in helicopter main rotor vibration such as multi-frequency sinusoids. The resulting filter removes outliers and reduces noise to a considerable extent when compared to a baseline noisy signal. FFT results show that the WRM filter is able to perform the noise removal while causing no change to the underlying frequencies in the ideal signal. The integer design space for the filter is multimodal with several local minima and the filter design problem presents a benchmark problem for testing integer programming algorithms. An exhaustive search of the design space ensures that all the local minimum points are found and the global minimum is identified. The 5-point median filter with weights [2 1 2 1 2] is found to be good choice for denoising vibration signals. The WRM filter can be used as a data preprocessor before subjecting the vibration signals to fault detection and isolation algorithm.