Model-based and Data-driven Approach for Machine Prognostics

Abstract

The severe competition in the market forced the industrial field to become cost-effective and reduce catastrophic failure. The maintenance activities including condition monitoring, fault diagnostic and prognostic are expected to overcome these issue. In maintenance work, the determination and prediction of final failure is the main objective. Therefore, prognostic part is become useful besides condition monitoring and fault diagnosis. Moreover, studies of prognostic approach are needed to be applied in industrial field. This paper proposes the application of particle filter (PF) method toward model-based prognosis approach and relevance vector machine (RVM) and logistic regression (LR) regarding data-driven prognosis approach. PF based sequential important sampling and resampling algorithm is employed to predict the trending data of low methane compressor and calculate the residual life of crack growth data of SPV50 steel. In terms of data-driven prognosis approach, RVM is combining with LR for failure degradation assessment. LR is used for calculate the failure degradation model of run-to-failure bearing data then RVM is employed to predict the final failure of individual bearing for case of simulated data and experimental data.