Cloud-Based Cad Parametrization and Image Recognition For Support of Engineering Design Using Numerical Simulations

In this work, an automated framework dedicated to design space exploration and design optimization studies is presented. The framework integrates a set of numerical simulation, computer-aided design, numerical optimization, post-processing, and data analytics tools using scripting capabilities. The tools used are open-source and freeware, and can be deployed on any platform. Two novel features are introduced in the workflow used. First, the use of a cloud-based parametric CAD tool that gives engineers and designers complete control over the geometry during the design loop. This feature allows users to deploy the design loop in any platform as the installation is not required. It also lets the designers interact with the parametric CAD model using a programmatic API. Introducing the CAD tool into the design loop has been traditionally a problem because most of the CAD applications run in Windows OS. In contrast, the simulation software runs in Unix-like OS. Furthermore, in traditional CAD tools is not possible to interact with the parametric model using a programmatic environment; they take all the inputs via a graphical user interface that cannot be controlled in an automatic design loop. The use of the cloud-based CAD tool allowed us to circumvent these problems. Secondly, the use of image recognition techniques, in particular, the SSIM index method, to drive the design study. By using this metric, it is possible to compare images instead of integral quantities. We can now design beforehand how the field will look like in a given location of the domain, and the design loop will try to find the best match for that qualitative metric. We demonstrate the capabilities and flexibility of the framework using computational fluid dynamics applications; however, the same workflow can be used with any numerical simulation tool (e.g., a structural solver or a spread-sheet) that is able to interact via a command-line interface or using scripting languages. We conduct design space exploration and design optimization studies using quantitative and qualitative metrics, and to reduce the high computing times and computational resources intrinsic to these kinds of studies, concurrent simulations and surrogate-based optimization are used.