In this work, meta-models for use in design optimization of low voltage motors are investigated. The idea is to develop an automated and efficient methodology for design optimization of a family of electric motors. A few widely adopted meta-modeling algorithms are examined with concerns of their accuracy and applicability for design optimization of the motors. Meta-model based optimization is conducted for a case of single motor with two objectives, and another case of a group of motors with shared design variables of cross-section dimensions and with an overall objective of total material cost. Meta-model based optimal designs are verified with that from real solver based optimization.
Computational expense for optimization simulations can be greatly reduced by using meta-models, especially for the family design case. Neural network models give the most satisfactory optimization result among all tested meta-models, in terms of accuracy and variety of the outcome designs in the objective space.
This work demonstrates great potential as well as challenge of meta-modeling technique for use in design optimization of industrial products and processes, where requirement on accuracy and reliability of the surrogate models being high.