Enhanced Morris for the extraction of significant parameters in high-dimensional design optimization

Abstract

The traditional Morris method cannot ensure the consistent extraction of significant parameters because its random sampling strategy often leads to an improper coverage of the input space, especially when the number of sample points is small. To overcome the drawbacks, we developed an improved sampling strategy to enhance the performance of Morris method based on Latin hypercube sampling (LHS) with the idea of Central composite design (CCD) and a fluctuant step to generate a more uniform points set. A comparison on the results of a numerical example obtained by the enhanced Morris, the traditional methods and CCS-Morris demonstrated that the former had a much better performance. The application of the enhanced Morris method in the high-dimensional optimization of a compressor shell with 15 design parameters aimed at increasing its natural frequencies and keeping its mass relatively light demonstrates that the method is effective and applicable in engineering design.