Injection molding is one of the most important plastic processes to mass-produce complex shaped parts efficiently with high requirements in part precision. Nevertheless, the setup process in injection molding can often require high experimental effort, since quality criteria like part dimensions or part weight are dependent on many influencing process settings in a complex manner. This effort is furthermore highly depending on the experience and process understanding of the individual machine operator and can expand up to multiple days or even weeks, depending on the part complexity and necessary tolerances. Reducing the setup time can deliver real benefit to the economic efficiency of a production. The following paper discusses the development and implementation of an injection molding cell combining automation techniques, data acquisition as well as quality measurement to be able to automatically create a process model for use in process setup and process optimization. This is achieved by using artificial neural networks as process models, which are trained by a new hybrid approach, which takes into account process real and simulated quality results. By utilizing virtual as well as real data, a sufficient model quality can be achieved with less experimental effort. To combine this methodology in an automated production cell, which is capable of performing the setup process autonomously, modern methods for machine data acquisition, data streaming, data processing and human interfacing are used. The described methodology has been implemented successfully in a demonstrator system, which is capable of performing the setup process autonomously using only a pre-calculated design of experiments from process simulation. The system was able to perform a setup process autonomously in about 3.5 hours, reducing the setup time and material usage. Since a process model is created during process setup, it can further be used for process optimization and adjustments at runtime.