An analytic approach to monitor main bearing health

摘要

In piston engines, failure of main bearings can lead to total engine failure causing huge financial as well as reputation costs for the organization. In this paper, an end-to-end analytic system, using data and domain, is described to develop a cumulative damage model to monitor the health of the main bearing using data obtained from engine lube oil analysis. The key outputs of the monitoring system are: (1) A multivariate baseline cumulative damage suffered by a 'typical' main bearing as a function of age. (2) Use of a 1-class support vector machine (SVM) to predict an impending engine failure because of a main bearing failure at least X days in advance. (3) Devise a ranking scheme for condition-based replacement of main bearings. The analytic system has been deployed for multiple railroad customers with a precision of over 90.