Exploration in Using the Weibull Distribution for Characterizing Trends in Bearing Failure Operational Changes

Remaining Useful Life (RUL) is critical to optimizing part life and reducing maintenance costs in a predictive maintenance strategy. Current methods of remaining useful life predictions are significantly dependent on operating conditions and time as input features. However, these features do not fully encompass the variability of real-world operating conditions and notably as the bearing nears failure. This work provides an improved failure representation by exploring the underlying data distribution parameters of a bearing failure dataset generated using the Purposeful Failure Methodology under varying operating conditions and then provides a comparison to the widely used NASA/IMS bearing run-to-failure dataset. Laboratory experiments utilized a bearing test stand to capture failure states for fatigue and contamination failure mode. The fatigue and contamination failure procession is compared to the failed bearings from the NASA Bearing dataset to examine similarities in the underlying data distribution between either dataset. A Weibull distribution is then fitted to both datasets. The resulting distributions exhibit similar trends, dependent on the damage stage. Based on the fitted parameters, a decreasing trend for the Weibull parameters was influenced by the changing speed in the engraving test case with similar trends to the NASA bearing dataset. The resulting understanding of the data distribution parameters will be used to improve the end of RUL calculation by describing the distribution fit that best determines the bearing life modification numbers.