Validation of Recurrent Failures Prediction Model Based on Underlying Distribution Function

SUMMARY & CONCLUSIONSThis is a sequel to the work published by the authors in RAMS‒2017, where they discussed regularization techniques for recurrent failure prediction using Kijima models. The main practical goal of such modeling is to predict the expected number of events (failures) in time, based on the underlying failure time distribution, which in turn, governs the random times between consecutive failures. The inverse problem of the model parameters estimation is ill‒posed and therefore can have several solutions, or the solutions can significantly depend on the accuracy of input data. Furthermore, the solution for the underlying distribution parameters can become in contradiction with that obtained by considering only the first failures. These aspects of the estimation problem are discussed in this paper by considering several typical examples, and a procedure to validate the obtained results is proposed.