Remaining Useful Life Prediction of Lubricating Oil With Small Samples

The remaining useful life (RUL) prediction of lubricating oil is essential for the preventive maintenance of machines, while the prediction accuracy has been severely limited by sparse and truncated data. Stochastic process modeling can provide a potential solution. However, two main challenges are encountered: 1) the nonlinear parameter estimation is prone to overfitting due to sparse data and 2) the lack of failure samples for threshold determination with truncated data. In the article, a novel RUL prediction model is developed based on the Wiener process and the oil degradation mechanism. Primarily, data augmentation is adopted to enhance data quantity for reliable nonlinear parameter estimation. Furthermore, the run-to-failure prediction based on the probability density function is performed to obtain the threshold with the truncated data. With the well-trained model, the RUL prediction is accomplished by updating the parameters and the thresholds with monitoring data. Furthermore, the prediction accuracy is validated with the oil data collected from both simulations and bench tests.