Product development can be accelerated by utilizing increasingly available data from manufacturing and service. Despite data availability, few methods can integrate design to qualify product systems and facilitate the design of a product’s next generation. This work introduces a Design from Manufacturing Data Mapping (DfMDM) framework and process to enable development of predictive analytics techniques to learn final system test results. Salient features of the predictive analytics include introduction of an optimal composition of simulation models to more accurately predict system test results from digital twin data while determining which simulation models are most significant. The approach is demonstrated by a case study that accounts for parametric and model uncertainty. These initial results show that this approach to optimally compose simulation models can reduce error in system test result predictions at early product development stages.