Input Smoothing for Faster Co-Simulation using FMI

Abstract

We present two technologies for speeding up co-simulationsunder the FMI standards. By smoothing the input signalsinside each FMU, the internal integrator may avoidre-initialization. This can significantly reduce the numberof model and Jacobian evaluations. To further help theintegrator we also propose a predictor compensationtechnique tailored to the input smoother. The main benefitof our technologies is the ease-of-use, requiring no modelmanipulations, nor any special co-simulation masteralgorithms. The technologies are implemented inDymola~2025x and validated with both an academic mechanicalmodel as well as thermo-fluid examples where we can observeperformance gains with factor up to 100, and often around5-10. One of these thermo-fluid examples is used in the\emph{OpenSCALING} research project to generate trainingdata for constructing surrogate models, for which the inputsmoothing is especially important to speed up the datasetcreation.