Model based control is generally seen as an indispensable tool to meet the coming challenges, but unfortunately models are difficult to determine. Available physical models are often too complex to be useful for control, and simplified models tend to be too imprecise. This paper proposes a mixed approach in which the model structure is determined by physical understanding, but the complexity of each block corresponds to its relevance for dynamical measurements and not to the physical complexity. Additionally, parameters are determined using standard measurements and optimization tools, and therefore do not rely on data from the producers of the components. This paper is focused on a nonlinear model for a turbo-charged Diesel engine with swirl valve and extends a previous work.