A Degradation Trend Dynamic Prediction Method for Rolling Bearing Health Status Evaluation

Rolling bearings as a key component of the high-end equipment, where there is a demand that the health status of rolling bearings should be evaluated timely and accurately for preventive maintenance. However, the non-smoothness and non-linearity of the collection signal make it challenging to characterize bearing degradation and extract degradation patterns. Additionally, the diversity of bearing degradation makes it difficult to classify and predict the degradation process. To address the appeal issues, this study is mainly carried out from two aspects: health index construction and degradation trend dynamic forecasting. Firstly, a healthy index with strong characteristics is conducted from the high-dimensional sensitive feature space. Secondly, the support vector data description (SVDD) method is used to identify the early degradation point. The dynamic threshold is later determined by the FPLR method for the historical data of the early degradation point to the current observation point. Finally, an autoregressive integrated moving average model (ARIMA) model is trained to dynamically predict future degradation trends. The aforementioned methods have been validated using publicly available datasets, and they hold significant importance in achieving intelligent warning and alarm systems for bearings, as well as improving the accuracy of remaining life prediction.