Advances in simulation-driven optimization and modeling

Abstract

Computer simulations are ubiquitous in contemporary engineering and science. In numerous fields, including mechanical engineering, civil engineering, electrical engineering, structural and aerospace engineering, automotive industry, oil industry, chemical engineering, ocean science and climate research to name just a few, simulation plays a critical role not only for verification purposes, but, more importantly in the design process itself. The complexity of structures and systems makes it analytically intractable, and it is thus extremely time-consuming and challenging to carry out any realistic design tasks, and in many cases, it is almost impossible to achieve any sensible design solutions under stringent constraints. These challenging tasks can be to optimally adjust the geometry and/or material parameters so that the system meets given performance requirements, or to calibrate the model parameters to make it fit given measurements, or to generate the optimal paths/routes for scheduling and planning tasks. In most cases, the interactions can be highly complex and multifold, and it is not easy or possible to isolate the processes of interest in the simplest, solvable form. For example, in the design of an electronic device, it is not just the isolated device to be designed that needs to be considered but also its – sometimes complex – interactions with the environment that affect the device’s performance. On the other hand, using accurate, realistic simulations allows the engineers to avoid costly prototyping and to realize the design closure with numerical models rather than through physical system measurements and prototype re-building. Furthermore, accurate simulations make it possible to analyze phenomena that could not be captured using simplistic theoretical models or too expensive or too time-consuming to be investigated through physical measurements. While high-fidelity numerical models can be very accurate, they tend to be computationally expensive. Simulation times of several hours, days, or weeks are not uncommon. In many cases, it may be a highly challenging task to just set up the model that takes into account all main, relevant system components and their interactions. One of the consequences is that a direct use of high-fidelity simulations in the optimization process may be prohibitive. The presence of massive computing resources is not always translated into computational speedup in practice, which is due to a growing demand for simulation