The optimal replacement problem for components with stochastic lives has an appealing solution based on the TTT-transform. The issue is revisited for components which are regularly inspected and where statistical uncertainties are taken into account by means of the method of predicted profile likelihood. The ideas are applied on crack growth data on a low pressure nozzle in a jet engine. It turns out that the standard method is not directly applicable and that the effect of uncertainties on the replacement times is not easy to predict.