Mechatronic systems such as vehicles, industrial machines, wind turbines, etc. as a system consist of one or more drivetrains which on their turn exist of components and subsystems. Drivetrains are key in various industrial applications and need to be reliable and have to assure proper operational performances. Nowadays increasing computational resources allow model-based scenario testing and design techniques. However, intricate drivetrains are plagued by nonlinearities and uncertainties that are hard to capture with solely physical laws. Traditional data-driven techniques on the other hand can have poor extrapolation capabilities, i.e. predictions outside their training region can become progressively worse. This project proposes a dynamic hybrid modeling formalism, by intimately merging physics-based with data-driven models that aims at improved accuracy, robustness and extrapolation capabilities. Research will be devoted to inserting these formalisms into virtual scenario testing for the detection of unwanted phenomena, as well as for optimal control strategy and design of drivetrains. Finally, this project delivers a generic toolbox that allows proper interfacing with the drivetrain experts and drivetrain systems to have optimal interplay with the developed algorithms.