A general data-driven algorithm for lifetime prediction

This paper introduces a general data-driven method to predict the lifetime of systems or components. Different from other methods, this algorithm doesn´t require the knowledge about the mechanism of object, thus applicable for various kinds of objects. The algorithm first analyzes the working curves to get a matrix which contains the historical information, and the degradation factors of all states can be calculated by the inverse or pseudo-inverse of this matrix. Then, with the degradation factors, the lifetime of objects can be predicted. The prediction methods are different for objects with or without future working plan. If there is a plan, the prediction algorithm can calculate the precise lifetime. But if there is no plan, the algorithm can still provide a range of the possible lifetime. The algorithm had been verified on a battery model, and the results meet the expected goals. In addition, this paper mainly considers the objects with discrete working states, and the continuous states will be studied in future papers.