In this work, a multi-fidelity framework for the simulation of small satellites is investigated. Taking into account the con-cept of digital twin, our work focuses on handling a constant stream of live data. Towards this end, current multi-fidelity modelling methods and low fidelity surrogate models for time series were surveyed. A multi-fidelity approach is used to combine a low fidelity surrogate model with a high fidelity model. As a high fidelity model, a previously investigated finite element model is assumed. As a low fidelity model, auto-regressive and recurrent neural network-based models are investigated. Through cokriging, the low fidelity data is corrected by the high fidelity data through a comprehensive correction, where the parameters are given through Gaussian processes in order to perform uncertainty quantification. As an application, the thermal simulation of a small satellite, and the use of this framework in conjunction with sparse teleme-try data is proposed. This online, statistical approach aims to provide a tool for performing fault detection.