Abstract

For the 2014 Prognostics and Health Management (PHM) Data Challenge Competition, the PHM Society proposed a problem surrounding risk prediction of engineering assets. We worked to address this problem by statistically analyzing the maintenance records, extracting key data features, and proposing an ensemble method for accurate prediction of imminent failure of assets. The data analysis of maintenance records provided two key pieces of information: 1) parts and part replacement reasons were able to be classified into corrective and scheduled maintenance actions, and 2) a linear relation was found between failure frequency and usage time. Based on this information, we proposed two risk-prediction methods, namely, a method based on part lifespan calculation and a method based on usage classification. Further work showed that the ensemble approach, which combined these two methods with a risk assignment formulation, provided more accurate risk prediction. The score predicted by the ensemble approach ranked in the second place in the 2014 PHM Data Challenge Competition.