Research On Rolling Bearing Health State Assessment Based On Modified HSMM

Rolling bearing is a widely used fundamental component of large rotating machinery,and it’s also a key unit to bear load and transmit motion for the equipment.The running condition of rolling bearing critically influences the performance of machine.However,rolling bearing always works in the environment with high speed and heavy load and endures alternating stress continuously,which can easily produce various defects of bearing working surface and cause the performance degradation,and eventually lead to the shutdown of the equipment.The performance degradation of rolling bearing is a gradual process of change,and based on the mastery for the degradation rule,the health state assessment and analysis can be realized.Then,the health state information of rolling bearing will be utilized maximally to provide guidance for the reliability maintenance,and then guarantee the safety and reliability for whole machine.This study takes rolling bearing as the research object,and researches on the health state assessment for rolling bearing.The main failure modes,failure mechanisms and fault evolution rules of rolling bearing are analyzed systematically,and multiple domain high-dimensional features are extracted to describe the performance degradation path of the rolling bearing.Furthermore,The feature selection and feature fusion methods are researched for the high dimensional multi-domain features set.Finally,a health state evaluation method based on the performance degradation process of the rolling bearing is studied.The main research contents are as follows:(1)Research on vibration mechanism analysis and multiple domain features extraction for rolling bearing.The main fault forms,the fault vibration mechanisms and the fault evolvement rules of rolling bearing are analyzed systematic.Then,the multiple domain high-dimensional features can be extracted from time domain,frequency domain and time-frequency domain respectively to reflect the performance degradation information of the rolling bearing,which provide a foundation for the later study of health state assessment.(2)Research on degenerate features fusion based on adaptive manifold learning algorithm.Aiming at the problem that the multiple domain features set has redundant and relevant components,three evaluation criteria of correlation,monotonicity and robustness are defined and the degenerate features selection method based on the Dezert-Smarandache theory(DSmT)for multi-criteria fusion is proposed,which is used to select the most optimized features that can better describe the performance degradation trend of rolling bearing.In order to eliminate the interference information in the optimized feature subset,a feature fusion method based on the local and global principal component analysis(LGPCA)manifold learning algorithm is researched.For the difficulty of neighborhood construction in LGPAC algorithm,an adaptive neighborhood selection method based on sample density and manifold curvature is proposed,which improves the feature fusion performance of LGPCA algorithm effectively.(3)Research on degradation prediction for the rolling bearing based on modified duration-dependent hidden semi-Markov model(MDD-HSMM).Since the time-invariance for key parameters in traditional hidden semi-Markov model affect its performance,the concepts of duration dependent health state transition probabilities and observation probabilities are proposed,and the new forward-backward algorithm and model parameters re-estimation algorithm are developed,which make MDD-HSMM describe the performance degradation process more realistically.Furthermore,the performance degradation state space model is created based on the MDD-HSMM,and the high-order particle filter method is applied to predict the degradation performance of rolling bearing.(4)Application research on degenerate feature fusion and health state assessment methods.With the perspective for modularized development,the software function modules of running condition monitoring,features extraction,features fusion and degradation prediction are designed and developed.Then,the rolling bearing health state assessment system is realized,which can verify the feasibility and practicability of the proposed algorithms in this study.