Rolling bearing is the core part of rotating mechanical equipment, so developing an effective remaining useful life prognostics method and alarming the impending fault for rolling bearing are of necessity to guarantee the reliable operation of mechanical equipment and schedule maintenance. The relevance vector machine is one of the substantially used methods for remaining useful life prognostics of rolling bearing. However, the accuracy generated by relevance vector machine drops rapidly in the long-term prognostics. To remedy this existing shortcoming of relevance vector machine, a novel hybrid method combining grey model, complete ensemble empirical mode decomposition and relevance vector machine are put forward. In the hybrid prognostics framework, the grey model is applied to gain a “raw” prediction result based on a trained model and produce an original error sequence. Subsequently, a new smoother error sequence reconstructed by complete ensemble empirical mode decomposition method is used to train relevance vector machine model, by which the future prediction error applied to correct the raw prediction results of grey model is projected. Ultimately, the online learning technique is used to implement dynamic updating of the “old” hybrid model, so that the remaining useful life of rolling bearing throughout the run-to-failure data set could be accurately predicted. The experimental results demonstrate the satisfactory prognostics performance.
Prediction of the Remaining Useful Life of Lithium-Ion Batteries Based on Empirical Mode Decomposition and Deep Neural Networks
