Abstract: The prediction of degradation trends in rotating machinery rolling bearings faces issues such as reliance on prior knowledge and low identification accuracy. To address these issues,a method that integrated Principal Component Analysis (PCA) with Random Forest (RF) and Gated Recurrent Unit (GRU) network for rolling bearing degradation trend prediction was proposed. Firstly,a set of high-dimensional features based on multivariate statistics was optimally selected. Dimensionality reduction and clustering were conducted by PCA to construct health indicators of the rolling bearing. Secondly,the health indicators were constructed to serve as basis,the RF model was introduced to fit the rolling bearings degradation curve. Finally,a PCA-RF-coordinated GRU network model of the rolling bearing degradation trend prediction was established to complete the rolling bearings status assessment. It is verified from experiment that the health indicators of the proposed method can effectively reflecting the degradation status of the rolling bearing,with the time trend of 0.999 1. Furthermore,it is shown that the PCA-RF-coordinated GRU model can accurately predict the degradation trends of the rolling bearing. The maximum root mean square errors for single-step and multi-step predictions on different datasets are 0.018 4 and 0.047 8,respectively.

Key words: rolling bearing, degradation trend prediction, principal component analysis, random forest, gated recurrent unit network, health indicators