Adaptive non-parametric methods of time-frequency analysis applied on marine propulsion systems

Abstract

Time-Frequency analysis is an invaluable tool for Rolls-Royce Marine as it helps scheduling maintenance when it is actually needed, rather than periodical appointments. A docked ship means reduced income so avoiding unnecessary maintenance is a step towards higher effectiveness.

With sensors placed in the thrusters of vessels, Rolls-Royce Marine extract as much information as possible from their machinery in order to give their customers feedback on the status of the machinery and inform them of faults that need attention.

Spectrograms have for now been the time-frequency tool of choice, but a need for an alternative that can achieve higher resolution has been expressed. Earlier work by the author has concluded that the reassignment spectrogram and synchrosqueezing spectrogram gives a better alternative to the spectrogram. Unfortunately these methods did not achieve the needed increase in resolution. That is why this report is focused on testing three adaptive non-parametric signal decomposition methods that have been shown to produce good results. These methods are empirical mode decomposition, local mean decomposition and Hilbert vibration decomposition.

Through testing these methods on various synthetic signals and sensor data from ship thrusters provided by Rolls-Royce Marine, it was concluded that none of the three methods had performance that led to recommending them for further use by Rolls-Royce Marine. Although initial results for Hilbert vibration decomposition seemed promising, further testing proved that the method introduced cross-term interference to such a degree that it can not be recommended for further testing.