This paper describes a methodology for the extrapolation of a single measured non-stationary time series to an expected long-term service history. Within the context of limited data availability, stochastic simulation is used to generate an ensemble of surrogate realizations from which expected long-term service histories can be derived. Two non-stationary stochastic simulation algorithms are implemented. Both simulation algorithms are compared using the transient response measured on hydroelectric turbine blades during startup. In both algorithms, an independent random phase shift is introduced in the analytic signal given by the Hilbert transform of each time series subcomponent. The subcomponents are obtained either by Empirical Mode Decomposition (EMD) or by Stationary Wavelet Decomposition (SWD). The simulated realizations will invariably include some inherent variations arising from the process itself, combined with epistemic uncertainty due to the assumptions made during modeling. To ensure the quality of the simulated realizations, the following quantitative criteria are used to compare the simulated ensemble to the reference data: cumulative energy, extreme value distribution and rainflow amplitude spectra.