An Intelligent Reliability Assessment and Prognosis of Rolling Bearings Using Adaptive Cyclostationary Blind Deconvolution and AdaBoost-Mixed Kernel Relevance Vector Machine

In this paper, a reliability assessment and prediction method based on bearing vibration signals is proposed, which combines Adaptive Cyclostationary Blind Deconvolution (ACYCBD) and AdaBoost-Mixed Kernel Relevance Vector Machine. Firstly, CYCBD parameters were optimized by the Ivy optimization algorithm to enhance the noise reduction effect, and then multidimensional features were extracted and dimensionalization was reduced by PaCMAP. Based on dimensionality reduction features, logistic regression was used to evaluate reliability, and AdaBoost-MKRVM was combined to predict reliability. The experimental results show that the mean absolute error (MAE) of the proposed method on the bearing life dataset of Xi’an Jiaotong University is 0.052, which is better than the traditional method, and provides a new idea for the performance prediction of rolling bearings.