Combining Relevance Vector Machines and exponential regression for bearing residual life estimation
Abstract
In this paper we present a new procedure for estimating the bearing Residual Useful Life (RUL) by combining data-driven and model-based techniques. Respectively, we resort to (i) Relevance Vector Machines (RVMs) for selecting a low number of significant basis functions, called Relevant Vectors (RVs), and (ii) exponential regression to compute and continuously update residual life estimations. The combination of these techniques is developed with reference to partially degraded thrust ball bearings and tested on real world vibration-based degradation data. On the case study considered, the proposed procedure outperforms other model-based methods, with the added value of an adequate representation of the uncertainty associated to the estimates of the quantification of the credibility of the results by the Prognostic Horizon (PH) metric. ⺠We use Relevance Vector Machines and exponential regression to compute residual life. ⺠The approach is applied to partially degraded thrust ball bearings. ⺠The approach is tested on real world vibration-based degradation data. ⺠We show that the proposed approach outperforms other model-based methods. ⺠The approach allows for the adequate representation of the estimates uncertainty.