The development of complex mechatronic systems requires the close collaboration of different disciplines, like mechanical engineering, electrical engineering, control engineering, and software engineering. To tackle the complexity of such systems, such a development is heavily based on models. Engineers use several models on different abstraction levels, for different purposes and with different view-points. Usually, a discipline-spanning system model is developed during the first, interdisciplinary system design phase. For the implementation phase, the disciplines use different models and tools to develop the discipline-specific aspects of the system. During such a model-based development, inconsistencies between the different discipline-specific models and the discipline-spanning system model are likely to occur, because changes to discipline-specific models may affect the discipline-spanning system model and models of other disciplines. These inconsistencies lead to increased development time and costs if they remain unresolved. Model transformation and synchronization are promising techniques to detect and resolve such inconsistencies. However, existing model synchronization solutions are not powerful enough to support the complex consistency relations of such an application scenario. In this thesis, we present a novel model synchronization technique that allows for synchronized models with multiple views and abstraction levels. To minimize the information loss and improve automation during the synchronization, it employs metrics to encode expert knowledge. The approach can be customized to allow different amounts of user interaction, from full automation to fine-grained manual decisions.