L-Kurtosis and Improved Complete Ensemble EMD in Early Fault Detection Under Variable Load and Speed

Abstract

In this work, we propose to follow the progression of different gearbox defects under the effect of variable load and speed. The non-stationary vibration signals are obtained by using a physical model of a spur gear transmission. In order to detect the presence of the fault characterized by transient signals which are usually masked by other vibration signals and noise. We can use the Improved Complete Ensemble empirical mode decomposition with adaptive noise (ICEEMDAN) to decompose the non-stationary vibration signals into many components that represent mechanical behaviour of the machine, transient component and noise. The ICEEMDAN method is based on the estimation of the local mean and the white noise is not used directly. this method eliminates the mode mixing introduced by EMD and reduces the amount of noise contained in the modes given by using EEMD and gives better results than EEMD. To analyze IMFs given by ICEEMDAN method we can use statistical methods like kurtosis which is very used to detect impulsion in the signal. In this work, we also use a statistical method, the L-Kurtosis, as an indicator to compare the IMFs given by ICEEMDAN, the results given by this indicator are compared to the results given by the Kurtosis.