The performance of small scale compressors mass-produced for domestic appliances is influenced by geometric manufacturing uncertainty which affects overall product performance and reliability. Precise manufacturing processes through tighter tolerances ensure high geometric accuracy and lower compressor performance spread, but it challenges high volume production capabilities and cost. Good understanding of geometric sensitivities is necessary for robust design and tolerance definition.

This paper presents the application of geometric Sensitivity Analysis (SA), Uncertainty Quantification (UQ) and robust optimization method to a small scale compressor. Geometric SA was carried out on parametric blade geometry and the key influential parameters were identified. The tip clearance was found to be the most influential parameter followed by a blade surface thickness. The tip section of the blade was more influential than hub section and the suction surface was found to be more important than the pressure surface. Findings from the SA were used to define parameters that were measured and controlled to ensure impeller quality during production. Within the parametric bounds studied, impeller chord did not feature as a critical parameter. The machining tool path was optimized accordingly and a 12% reduction in cutting time was achieved. The numerical sensitivities were also compared with experimental data and a trend-level agreement was seen. Meta-models for aerodynamic performance were built using DoE generated geometries which were used to perform manufacturing UQ using a Monte Carlo estimator. Robust design optimization was carried out using stochastic optimization algorithm coupled with the meta-model based Monte-Carlo simulator. This framework was used to choose a robust nominal shape that achieved 18% lower standard deviation in stage pressure rise. The predicted performance spread was compared with production data and a satisfactory agreement was seen.

This content is only available via PDF.
You do not currently have access to this content.