Recent advances in sensing, data analytics, and manufacturing technologies (e.g., 3-D printing, soft robotics, nanotechnologies, etc.) provide the potential to produce highly customized products by allowing flexible system design, endless device configurations, and unprecedented information flows. These opportunities also increase the complexity of controlling such systems optimally, which typically requires fast exploration of an increasingly large number of alternative operation strategies. Simulation and stochastic models have been particularly successful to support control and optimization of production systems, and methods have been developed to exploit them separately. Herein, we argue that the simultaneous use of these models can allow for better control and optimization by balancing the simulation accuracy, and related high computational costs, with the computational efficiency and lower accuracy of stochastic models. In this article, we assume that high fidelity models have higher accuracy and computational costs, and we present a novel multifidelity approach, which utilizes several models at different levels of fidelity to efficiently and effectively estimate and optimize the performance of asynchronous serial production lines with machines suffering from multiple failure types. Experimental results show that the multifidelity approach leads to better estimations, requiring less computational effort for optimization compared with the use of only high fidelity simulations.