The recent trend towards using simulation models with real-time data as digital twins is rapidly increasing in industry and research. In this presentation, a digital framework supporting real-time geometrical quality control of assemblies will be presented. The approach builds on a simulation framework, where non-rigid, statistical variation simulation forms the basic model. The simulation uses scan data on part level as input and predict the geometrical deviation from nominal values on assembly level. For each individual assembly, a digital twin is built and joining process parameters, such as joining sequence and locator positions, are optimized to reach the maximum geometric quality for each assembly. In the concept, also selective assembly, i.e. matching of parts, is included. The joining process is in focus, and different joining methods are considered in the digital twin. For the moment, spot welding or riveting are covered, but on-going research is focusing on including more complex simulations covering the effect from heat and continuous welding. This is challenging, both with respect to accuracy and simulation time. The approach is implemented in RD&T, which is a software for variation simulation. The non-rigid package is based on finite element analysis, combined with statistical variation simulation using monte carlo simulation and the method of influence of coefficient approach. RD&T is a spin-off from academia but is also commercially available and used by a large number of companies, especially in automotive and aerospace industry. It is also used as a test bench for research projects at Chalmers University of Technology. The presentation will be concluded with numerical results from case studies from automotive industry, showing the potential for improvement of the geometrical quality of assembled products. Challenges related to the concept will also be discussed. Those cover areas such as quality of inspection data used as input to the digital twin, as well as requirements on simulation time to reach a real-time digital twin approach.