RUL Prediction of Rolling Bearings Based on Crested Porcupine Optimization Algorithm Optimized CNN-BiGRU-Attention Neural Network

A traditional rolling bearing life prediction model often suffers from low prediction accuracy and weak robustness. To further enhance the accuracy and robustness of the prediction model, a rolling bearing remaining useful life (RUL) prediction model is proposed, integrating four modules: the Crested Porcupine Optimization (CPO), Convolutional Neural Network (CNN), Bidirectional Gated Recurrent Unit (BiGRU), and Attention mechanism. Firstly, to accurately describe the bearing degradation process, a Health Indicator (HI) for the bearing is constructed using Analytic Hierarchy Process (AHP) and Principal Component Analysis (PCA). Then, fault points on the HI curve are detected using Cumulative Sum (CUSUM), and the steady-state data is removed. Finally, the data after the fault points is input into the CPO-CNN-BiGRU-Attention model for RUL prediction. Through comparative validation with CPO-BiGRU-Attention, CPO-CNN-Attention, CPO-CNN-BiGRU, and CNN-BiGRU-Attention models, it is demonstrated that the proposed RUL prediction method achieves the lowest prediction error and the highest prediction accuracy.