Residual useful life prediction of large-size low-speed slewing bearings - a data driven method

In this paper, a data driven method based on Ensemble Empirical Mode Decomposition-Principal Component Analysis (EEMD-PCA) and Least Square-Support Vector Machine (LS-SVM) is proposed to achieve residual useful life (RUL) prediction for slewing bearings. Firstly, life-cycle vibration signals are acquired and divided into several signal segments, and EEMD is then utilized to decompose each segment into Intrinsic Mode Functions (IMFs). Afterwards, PCA is introduced to illustrate the trends of each IMF across the life cycle, and some of the IMFs that contribute to reveal the performance degradation process of the slewing bearing are selected to reconstruct and de-noise the vibration signals. After that, continuous squared prediction error (C-SPE) and its features are presented as performance degradation indicators. Finally, an RUL prediction model is built on the basis of the indicators using LS-SVM. The results of a life-cycle experiment show that the C-SPE of the de-noised vibration signals can precisely explain the performance degradation process of the tested slewing bearing and that the established RUL prediction model is close to practice. Besides, a comparison study shows that the C-SPE based RUL prediction model is more efficient and accurate than the signal based model. Therefore, the proposed method ensures high-reliability slewing bearing prognostics.