Remaining useful life prediction for multi-component systems with hidden dependencies

Abstract

How can we predict the remaining useful life (RUL) of a dynamic system subject to multiple dependent degradations? Is it possible to address the above problem when the degradation information is obtained indirectly? According to a new type of state space-based model, we mainly develop an online RUL prediction method for the above system. In this model, the dependencies among different degradations can be reflected in a diffusion coefficient matrix. Considering that some industrial systems like blast furnaces are usually equipped with multi-sensors, an efficient information fusion strategy also plays an important role in predicting the RUL. Based on multi-dimensional observations, the hidden degradation states are identified through the sequential Kalman filtering. Meanwhile, the unknown parameters in the model are updated iteratively by the expectation maximization (EM) algorithm. At last, the RUL distributions are simulated through the Monte Carlo method, in which three types of failure structures with regard to the degradations are considered. The effectiveness of the proposed method is fully verified by a numerical example as well as a case study about the blast furnace.