Enhancing Degradation Models of Mechanical Systems Through Uncertainty Quantification

Abstract Over the years, various techniques have been developed to construct remaining useful life (RUL) models for reliability analysis based on run-to-failure experiments for rolling element bearings. The naturally produced experimental data from open-source repositories, XJTU-SY and PRONOSTIA ball-bearing datasets, were used in this study. The PRONOSTIA dataset has no records for its location of damage, and knowledge of the bearings’ damaged components will benefit studies requiring it. Both datasets are testaments to fault commencement and complete failure occurring at different times for identical bearings running under the same operational conditions. Hence, a bearing will have multiple points where incipient fault points (IFs) and end-of-life points (EOLs) could occur. Thus, attributing all this information to a singular degradation model is essential. This study answers the research question: Can quantification of uncertainty enhance degradation models for rolling element bearings? The objectives of this paper are to (1) identify the unknown damage locations, (2) quantify the uncertainty of the IFs and EOLs, and (3) determine the degradation model considering uncertainty. Due to limited data from both datasets, the study employs the maximum entropy principle (MaxEnt) to define uncertainty for both the IFs and EOLs. This is achieved by plotting probability density functions (PDFs) such that both the IFs and EOLs transcend as random variables (RV) instead of deterministic parameters. MaxEnt is further used to construct a degradation model by connecting the PDF, cumulative distribution function, and reliability model. An approach based on uncertainty quantification was proposed to determine a degradation model of ball bearings. The approach is shown to be generalizable to other engineering systems.