In the field of engineering, surrogate models represent a promising alternative to conventional simulation approaches. Especially due to the fast computation time, the predictions can be used, for example, to validate concepts early in the development process or to provide engineers with a real-time user interface that can support the design of future systems. The development process of surrogate models involves several steps - sampling, training and validation. In the field of simulation-based surrogate models, training data must be generated first by performing expensive high-fidelity simulations. In order to keep the simulation effort as low as possible, design of experiments (DoE) methods are applied. Depending on the type of surrogate model - physical, data-based or hybrid - and the type of problem, a varying number of training simulations is required. In the training step, the model parameters are then optimized to provide the best possible representation of the desired solution. Before the surrogate model can be used for prediction, validation must be performed with an independent test set that was not used for training. Surrogate modeling is a universally applicable method and can be profitably used in many areas. This includes any kind of FEM, MBS, CFD and other simulations. Surrogate models can approximate simple scalars, but also time series data or even deformations of FE meshes. Several examples of different surrogate modeling approaches are shown below.