Analysis of crack growth with robust, distribution-free estimators and tests for non-stationary autoregressive processes (Q2442687)

In order to predict crack growth, it is necessary to predict the direction and magnitude of crack growth at each iteration. This operation required a model for crack growth. One of the best models is the stochastic model. In this paper, the authors are adding a stochastic error term, then they model the crack growth by a non-stationary autoregressive process with Lévy-type errors. The drift parameter of the process is estimated by using the regression depth approach. The authors prove the consistency property of that estimator under quite general assumption on the error distribution. They also use a degenerated U-statistic and establish tests for general hypotheses about the drift parameter, based on an extension of the depth notion to simplicial depth. An empirical part is conducted by using simulation of AR(1) processes with different error distributions (Gumbel, Mixture of Poisson and normal, Normal Frechet), to examine the quality of the constructed test and some other aspects related with.