Complex physics-based models are a common way to run experiments on computer hardware. However, high-fidelity simulation models require a substantial investment of computation time. This is prohibitively expensive for routine tasks such as optimization, sensitivity analysis, and design space exploration. Metamodels are data-driven approximations that mimic the behavior of the simulation model as closely as possible while being computationally cheap(er) to evaluate. Metamodel-assisted optimization is a popular approach to expedite design optimization in engineering. However, multi-objective optimization remains expensive as it wants to find the full Pareto front, while the decision-maker (DM) is only interested in a small subset from an infinite number of possible (optimal) solutions. The final solution selected by the DM will depend on his or her preferences. The goal of this project is to add user preferences to multi-objective optimization to scale up to many objectives in a data-efficient manner. The key research questions in this regard are: 1) Learning the user preferences interactively: During the optimization process, preference information can be obtained by interacting with the DM. This way, convergence to the truly most preferred solution(s) might be accomplished without wasting computational effort, 2) Imperfect knowledge: The preferences of the DM have a complex noise structure due to cognitive overhead and undesired biases with a temporal component.