Experiment Driven Local Optimization (EDLO) With an Application to Additive Manufacturing
The rise of the use of additive manufacturing processes by engineering and research enterprises has greatly increased opportunities for decision makers to quickly evaluate complex geometries and components throughout the design process. This capability makes it possible to explore design trade-off for which computational or analytical models are not readily available, or are impractical to obtain given available time and resources. However, this often means that decision makers are forced to rely primarily on physical experiments for design data, and this greatly limits opportunities for the application of design optimization techniques. To meet this challenge, a new so-called “online” surrogate based optimization approach, called Experiment Driven Local Optimization (EDLO), has been developed. This approach is focused specifically on using approximation techniques and real-time online surrogate training to solve design optimization problems where the system objective function must be evaluated using physical experiments. As a result, this approach is ideally suited to design problems focused on components that are fabricated using additive manufacturing. This approach is capable of mitigating the effects of experimental uncertainty (or noise) in a design objective function, does not require close coordination between the objective function evaluations and the optimizer, and requires as few physical experiments as possible. These capabilities have been demonstrated through the use of several numerical test problems and also through a 3D printed compliant mechanism design problem. The results produced by the EDLO approach are encouraging and show that this new technique has the potential to move decision makers working with physical experiments and additive manufacturing away from engineering intuition and towards the greater potential of design optimization techniques.