In particular, we specialize in the development and industrial application of gradient-based optimization algorithms, with emphasis on the adjoint method. Part of the developed adjoint software has been made publicly available since OpenFOAM v1906. Following our long experience with OpenFOAM, we can undertake the solution of shape or topology optimization problems using existing software or the development and implementation of new adjoint-based optimization algorithms within OpenFOAM. The flow model may include steady or unsteady, incompressible or compressible stationary and rotating flows, heat transfer or buoyancy effects, etc. In addition to a large collection of objective functions already supported by our software, new objective functions can be implemented and integrated into an optimization loop, on demand. Moreover, the setup of the flow, adjoint and optimization problems can be tailored to specific industrial needs, targeting an automated, easy-to-use everyday analysis and optimization tool.
GPU accelerated software can be one to two orders of magnitude faster than the corresponding CPU codes depending on the scientific field of interest and the special characteristics of the software. FOSS experienced GPU programmers can port existing CPU codes to GPUs.
The GPU specialized team can also provide consultancy for code optimization and software tuning in order to achieve the maximum performance according to the hardware in hand and its computing capability. Possible areas of improvement include, but are not limited to, optimal GPU memory accessing, customized variables, etc.
We have developed global, evolutionary algorithms-based optimization solutions suitable for engineering problems from any discipline since the optimizer is agnostic to the evaluation tool. The software makes extensive use of response surface methods and artificial intelligence to speed-up the optimization process.
Finally, FOSS can help you choose the best combination of global and local optimization methods for obtaining an optimal design in the shortest time.
Our automated design processes :
- are suitable for complex engineering problems,
- are supported by smart sampling techniques,
- include various surrogate models (RBF, Kriging, etc),
- use state-of-the art bio-inspired optimization algorithms,
- are compatible with any in-house and commercial software.
Predict the quality of new designs based on the experience of previous designs using deep neural networks and machine learning techniques