Time-weighted kernel density for gearbox residual life prediction

With improvements in industrial automation, the reliability of the gearbox, a key transmission device, has become increasingly crucial for the stable operation of an entire operating system. However, predicting the remaining useful life of the gearbox is challenging because of complex working environments and dynamic load changes. Several existing methods assume an inaccurate model structure and parameter estimation during life prediction, owing to the limited availability of similar fault sample data. In this study, we analyse the influence of kernel density estimation (KDE) based on time-varying distribution on the results of residual useful life prediction, considering the characteristics of such systems and the problems faced by current research methods. First, a time-varying KDE model with an incremental distribution of degradation features is established, and the influence of sample timing on KDE is introduced. Second, the exponential weighted moving average method is employed to predict the degraded samples, and recursive update was employed to reduce unnecessary double calculations during the estimation of the time-varying weight kernel density in the system operation process. Finally, the adaptability and effectiveness of the proposed method are verified using actual collected gearbox data. Research results indicate that the remaining useful life prediction outcomes of the method proposed in this paper are superior to those of the DGN model and the Ensemble model, as evidenced by its lower RMSE and MAE values.