“Computer optimization eliminates trial and error in using CFD and combustion simulation to design engines,” said Kelly Senecal, Vice President of Convergent Science Inc. (CSI). Automated optimization means letting computers do the hard work of getting the right design. This means balancing a mix of design parameters, such as spray-injection timing, injection-rate profile, number of spray pulses, or spark timing. It even means adjusting cylinder and piston geometry for best performance. While human designers set what parameters to vary and what they want for performance, computers do the nitty-gritty detailed work of getting it just right.
What makes CSI's Congo optimization different from others is its Genetic Algorithm. Senecal says this gives designers more confidence that they have found the best solution. Engineers could run a large number of simulations varying one factor at a time (trial and error.) This is time consuming and yields uncertain results. “You could also optimize using design of experiments (DOE), but that gives a local optimum,” explains Senecal. Such methods could neglect parameter interaction or provide solutions only inside the boundary conditions set by the DOE. On the other hand, global optimization methods such as Genetic Algorithms inherently include interaction effects. They also tend to converge to a global optimum for multi-modal functions with many local extrema, according to Senecal.
“Genetic Algorithms can think ‘outside the box’ and provide solutions designers may not have considered,” he explains.
How do Genetic Algorithms work? The key is to develop an output merit function within Congo. This function includes the parameters to optimize, such as engine fuel consumption. It also imposes constraints, such as maximum allowable engine emissions. A set of input parameters defines an individual solution. Multiple solutions define a population, the size of which the user defines. Once set, Congo exercises Converge CFD multiple times to fill out the population. Congo retains the best solution of a single iteration while it generates new solutions to fill out the population set for each iteration. As the computation iterates, individual solutions become more similar to the fittest and converge to a single solution (within defined criteria).
“Genetic Algorithm is survival of the fittest. It is not new. What we have done here is apply it to engineering of internal-combustion engines,” Senecal said.
Congo works with the company’s Converge software, and now offers it as part of its solution set to customers. According to Senecal, Congo’s automatic optimization would not be feasible without Converge CFD’s ability to create and refine computational meshes automatically as well.