Condition-Based Bearing Inventory Management Using NASA/IMS Run-to-Failure Data

Abstract

This paper develops and evaluates an integrated framework for condition-based inventory management of rollingelement bearings using run-to-failure vibration data. Starting from raw acceleration snapshots, a scalar health indicator based on root-mean-square (RMS) vibration is constructed, smoothed and normalized to obtain monotone degradation trajectories for individual bearings. A Gamma stochastic process is fitted to the positive increments of this health indicator, allowing Monte Carlo simulation of time-to-failure and remaining useful life distributions as functions of the current health level. These reliability measures are then embedded in a periodic-review inventory model formulated as a Markov decision process for a single spare-part class. The state captures the on-hand inventory; demand is driven by degradation-induced failures, and the objective is to minimize the expected discounted sum of ordering, holding and stockout costs. Numerical experiments based on the NASA/IMS bearing data illustrate how the proposed approach can support leaner spare-part policies while maintaining a low risk of stockouts and highlight the impact of degradation parameters and cost assumptions on the optimal ordering strategy.