Letting swarm intelligence adapt a pretty-good base design into multiple optimal variants.
To achieve fit for different applications of its products, this equipment supplier needed to increase the number of variants of a complex mechanical component.
The variants were all derived from the same base design, which had already been incrementally improved by hand to achieve very high performance while meeting very strict target conditions for flow rate and efficiency. The challenge was to modify the base design so that each target variant would achieve a different target flow rate, yet with the best-possible performance.
After analyzing the problem and remapping the manual simulation-based design process, an engineering optimization strategy was created. Instead of relying on slow iterations with expensive, high-fidelity simulation models, the new approach relied on low-fidelity models and on a swarm-intelligence optimization algorithm to enable a wide, robust exploration of the design space. Keeping the engineer in the loop made it possible to progressively exclude suboptimal parameter ranges and progressively fine-tune the algorithm.
The result of the optimization strategy was a set of five optimal component variants that were validated using high-fidelity simulations and test rig measurements.
The benefit of employing an engineering optimization strategy was that, after the first variant had been optimized, the strategy was applied in a fully-automated for the rest of the variants, essentially providing a a triple digit speed boost compared to the classical manual iterative trial-and-error simulation-based approach.
The resulting Python code optRATIO is proprietary, but the approach was made public with the paper “Design of Radial Turbine Meridional Profiles using Particle Swarm Optimization” at the 2nd International Conference on Engineering Optimization (EngOpt2010).
