We present two technologies for speeding up co-simulations under the FMI standards. By smoothing the input signals inside each FMU, the internal integrator may avoid re-initialization. This can significantly reduce the number of model and Jacobian evaluations. To further help the integrator we also propose a predictor compensation technique tailored to the input smoother. The main benefit of our technologies is the ease-of-use, requiring no model manipulations, nor any special co-simulation master algorithms. The technologies are implemented in Dymola~2025x and validated with both an academic mechanical model as well as thermo-fluid examples where we can observe performance gains with factor up to 100, and often around 5-10. One of these thermo-fluid examples is used in the \emph{OpenSCALING} research project to generate training data for constructing surrogate models, for which the input smoothing is especially important to speed up the dataset creation.

This session is chaired by