Restraining EMD end effect of vibration signal based on Homotopy Least Squares Support Vector Double Regression

Aiming at the end effect in Empirical Mode Decomposition (EMD) of vibration signals, Shape-Preserving Piecewise Cubic Spline (SPPCS) interpolation method and Homotopy Least Squares Support Vector Double-Regression (HLSSVDR) algorithm is proposed in this paper which can restrain the end effect. In order to obtain valid upper and lower envelopes, based on the studies of end effect mechanism and the existing research results, the SPPCS used to eliminate the fitting overshoot/undershoot problems by structuring the upper and lower envelope curves of the signal. Then the HLSSVDR algorithm is introduced to predict and replace the left and right values at both ends of the mean values of the upper and lower envelopes. The proposed method is analyzed and tested using the actual rolling bearing vibration signals. The experimental results indicate that the proposed method can effectively inhibit the fitting overshoot and undershoot, as well as the end effect, with higher accuracy and less distortion decomposition.