Titelaufnahme

This article presents a new method to test on whether a parametric model is capable of describing data properly. It also introduces a simple procedure to generate simultaneous prediction bands based on independent copies of a process. The performance of these prediction bands, e.g. in a leave-one-out cross-validation, will also be used as another indication of whether data is modeled properly. Both methods are applied to data from fatigue experiments on prestressed concrete beam girders. These experiments highlight a couple of different influences on the fatigue of such girders, namely the so-called cable factor and the deflection force. Both effects are incorporated into different load sharing models for component failures which then are compared and used for predicting these failure times.