Inverse engineering: A methodology for learning models to support engineering design

Synthesis tasks in engineering design are currently performed by iteratively using computer-based analysis simulators in a generate-and-test fashion. In the inverse engineering methodology machine learning techniques are used to learn bi-directional models which can directly provide synthesis support. This greatly reduces design iterations. Inverse engineering provides decision-making support during the early stages of design. The authors focus on supporting early product design in parameterized domains. They present the three main phases of the inverse engineering methodology and briefly describe the model formation and selection phases. The emphasis is on the final model-utilization phase where the models learned are used by the design engineer to efficiently explore the tradeoffs in the design space in an interactive fashion. Two examples from the early design of combustion chambers for diesel engines are presented