Dynamic Process Chain Control for Compressor Design Optimization

Kurzfassung

Modern optimization workflows involve interdependent quantities of interest (QoIs) across multiple disciplines and operating points. The scheduling of subordinate process chains strongly influences optimizer convergence and computational cost. This study presents an automated approach that emulates expert decision-making to select which process chains to execute, thereby avoiding unnecessary evaluations and reducing wall time.

A dynamic QoI-prioritization method by dynamic process chain control (DPCC) is introduced to enable runtime selection of the next process chain. By linking progress-related parameters to each chain, their importance is assessed using derived criteria such as constraint violation, surrogate reliability, predicted improvement, and information-to-cost ratio. The framework integrates seamlessly with common surrogate-based and gradient-free optimizers.

The method is demonstrated through two test cases: a multi-operating-point optimization of a quasi-3D compressor cascade and an aero-structural optimization of a low-pressure compressor rotor blade. To account for optimizer stochasticity, each case was repeatedly solved from identical initial designs. Results are statistically compared to a static reference executing the full process chain at every iteration.

Across both problems, the DPCC strategy consistently yields decisions aligned with engineering intuition while achieving significant reductions in computational time, depending on the problem, without compromising solution quality, demonstrating its potential as a practical approach for accelerating multidisciplinary compressor optimizations.