Augmented reality devices can be used by engineers and scientists to perform numerical simulations for different physical systems. Using augmented reality glasses for presenting simulation results, in the form of virtual objects, improves interactivity and quality of user experience (QoE). However, despite availability of efficient approaches for fast approximate computations, such as the reduced basis method (RBM), it is not feasible to preload all reduced models for different simulations and sets of parameters in advance. Therefore, an approach is presented, that allows to detect real-world objects, map specific tags to corresponding simulations and to efficiently use available storage. By analyzing the environmental sensor data and estimating the required quality, determined by user perception, only relevant reduced bases are loaded to the internal storage of mobile device. The proposed system periodically obtains solution for the given simulation in accordance with changed parameters and quality demands, either by loading updated reduced model from server or performing the computation of the full problem directly on the mobile device in case of network failure. Simulation results are visualized by means of hologram, overlaid onto the detected object with exact position and pose. Evaluation results show that solving the full numerical problem on mobile device is feasible for dimensions up to 32x32, whereas for higher quality constraints the latency of requesting the reduced model from server can be improved by up to 10 times by performing preliminary availability check and starting computation in advance.