Doubly Stochastic Cumulative Damage Model for RUL Prediction of HDDs in Uncertain Operating Environments

Degradation modeling and remaining useful life (RUL) prediction for hard disk drives are crucial for exploring the failure mechanism, evaluating the design reliability and improving the design. Motivated by the fact that magnetic head degradation is the most prevailing failure mode, a variety of methods for head degradation problem have been proposed. For example, non-homogeneous Poisson processes were developed to deal with such problems. However, a general non-homogeneous Poisson process assumes that the mean cumulative function is deterministic. This assumption cannot be satisfied when the degradation process contains stochastic dynamics in uncertain operating environments. To solve this problem, this article proposes a non-homogeneous Poisson process with a stochastic mean cumulative function that uses Gaussian noise terms to capture the stochastic effects. Then, a nonlinear state-space model is built using the proposed model to track the dynamic degradation process and predict the RUL of magnetic head under uncertain operating conditions. To estimate and update the model parameters adaptively, a stochastic filter and a maximum likelihood estimation (smoother) based on the expectation-maximization algorithm is derived. Finally, a real case study is used to illustrate the effectiveness of the proposed method.