ESI announces pilot version of Adjoint Solver for design optimization

Numerical optimization has proved to be an essential technology to achieve some ever-increasing requirements of the automotive sector. Optimizing the shape of a vehicle allows industrials to achieve greater fuel efficiency while using innovative, lighter materials. Numerical simulation can be implemented at each and every step of product design to predict the aerodynamics, aero-acoustics or even the aero-thermal performance levels resulting from a specific product design.

Standard CFD solvers versus adjoint-based optimization method

Computational fluid dynamics (CFD) analyses are very demanding at an industrial level because of the number and duration of the computations they require. Further-more, the genetic algorithms traditionally used in structural optimization are not suited for CFD optimization. Gradient-based methods, which converge to the optimum in many fewer iterations, require the computation of the cost function’s gradient, which is not provided by standard industrial CFD solvers.

Addressing these problems, the adjoint-based optimization method has progressively emerged as a preferred method for CFD calculations. This method has been recognized as most reliable by major aerospace companies over the past decade and is now being adopted by the automotive sector.

A synergy of programming technological advances now allows the use of the adjoint-based optimization method on an industrial scale

In an effort to make this method accessible to various industries, and to ease up code maintenance, ESI has recently worked towards an independent  discrete adjoint solver that takes the form of a dynamic library coded in advanced Fortran 90.

This library aims to abstract all the derivative machinery required for assembling and solving the adjoint system, irrespective of any discretization scheme of the application code. It has been fully validated within ACE+ Suite, ESI’s multiphysics software. In support of this project, ESI and Intel have engaged in a fruitful collaboration over the last 2 years, bringing the Intel F90 compiler to a unique level of performance with respect to most advanced F90 features.

”Intel was delighted to help ESI build their advanced solver with the Intel Fortran Compiler. This is a nice illustration of benefits of Intel’s long-standing enabling engagement in code optimization with ESI. This collaboration gave ESI the added performance they wanted and helped improve the Intel compiler for all customers,” says Laurent Duhem, Software Application Engineer at Intel.