Main bearings deployed in the field may fail after 6 to 10 years of operation, leading to costly replacements ranging from $150,000 to $300,000+. Multiple main bearing failures in a budget year can bleed heavily into profit margins, because in addition to the cost of the replacement bearing, there’s a loss of revenue incurred from unplanned downtime. Therefore, the industry has been looking for ways to predict main bearing failures and reduce the failure rates within their fleet.
Traditionally, operators are reliant on corrective maintenance practices, using data from temperature and vibration sensors to identify main bearing failures. However, the reliability of these sensors is questionable and by the time a noticeable difference is detected in temperature or vibration, it is already too late to exercise life extension actions, which could potentially increase the remaining useful life of the bearing. Conventional computational models identify wear rates by conducting experiments to understand wear coefficient and macroscale inputs for a Finite Element Analysis (FEA) model or empirical equations, which generate a deterministic bearing life. However, tribology-related failures are very complex and probabilistic in nature.