GENETIC ALGORITHMS TO DETERMINE THE OPTIMAL PARAMETERS OF AN ENSEMBLE LOCAL MEAN DECOMPOSITION

Resumo

An optimization method for an ensemble local mean decomposition (ELMD)

parameters selection was proposed by using evolutionary algorithms. ELMD is an adaptive, non-
stationary and non-linear signal processing method based on the addition of different white

noises to the target signal to be decomposed into a local mean decomposition (LMD) in order to
solve its large mode mixing problem. Even though it has satisfactory performance for fault
diagnosis of rotating machinery and reduced maintenance costs, the execution of this technique
depends heavily on the correct choice of the parameters of its model. Although it was proposed
to optimize the selection of these parameters through relative root-mean-square error (RRMSE)
and signal-to-noise ratio (SNR), the optimized ensemble local mean decomposition (OELMD)
based its selection by testing several values for amplitude, noise bandwidth and ensemble trials in
order to obtain the optimum value. However, even with excellent signal processing results, this
technique can raise computational costs and become prohibitive, especially in real-time analysis.
Thus, this work also proposed an optimized ELMD, but in this occasion using the inclusion of
genetic algorithms. The effectiveness of the proposed method was evaluated using synthetic
signals, which was used by several authors for such purposes. Despite presenting inferior results
to OELMD in avoiding mode mixing problem, the suggested algorithm obtained better results
regarding the signal processing time.