Methodology for Automated Tuning of Simulation Models for Correlation with Experimental Data

In this paper a practical methodology for automated tuning of simulation models is introduced, which is widely and successfully adapted in IAV. For this, stochastic optimization algorithms (like Genetic Algorithms or Particle Swarm Optimization), and appropriate algorithms for optimization tasks with very long computation time (e.g. Adaptive Surrogate-Model Optimization or Adaptive Hybrid Strategies) are used in combination with commercial and internal simulation tools. Often it is necessary to evaluate several contradictory objectives at the same time which leads to multi-criterion optimization. Effective post processing methods (mathematical decision aids) are used to select the best compromises for the problem.

As a practical example, this automated tuning methodology is applied to an engine performance simulation model developed in GT-Power. Procedure of multi-criterion optimization for co-relation of output parameters like rate of heat release, burn duration, 90% mass fraction burned etc. is explained in detail. It is observed that, time required for simulation model tuning is reduced by up to 75% w.r.t. conventional methods of model tuning. A good co-relation w.r.t. experimental data is achieved even for cases with lots of parameters and multiple operation points.