Recently, designing mechatronic systems has become more and more dependent on models that are used to predict performance in a virtual environment, and the models involved are becoming increasingly more complex multiphysical systems. Instead of spending much time modeling increasingly detailed physical models, uncertainties can be explicitly considered to model the lack of knowledge. The mismatch between real-life experiments and model simulations due to parametric uncertainties can be quantified using likelihood estimation and Monte Carlo sampling techniques for propagation. In this paper, we attempt to significantly accelerate the process using polynomial chaos expansions for propagation and a genetic algorithm to maximize likelihood. The soundness of this approach is demonstrated on a wet friction clutch system. The results show that the method has a strong potential for scalability with respect to the number of uncertain parameters.