ScholarWorks@성균관대학교

초록

Manufacturing tolerances are a primary source of performance degradation in electric motors, making robust design optimization (RDO) essential. However, conventional RDO methods that integrate tolerance analysis directly into the optimization loop suffer from high computational costs due to the vast number of required Finite Element Analysis (FEA) simulations. To address this challenge, this paper proposes a computationally efficient and experimentally validated framework for robust motor design. The proposed methodology decouples the initial multi-objective optimization from the subsequent robustness evaluation. First, a set of Pareto-optimal solutions is obtained using the NSGA-II algorithm. A deep learning (DL) surrogate model is then trained on the data generated during this optimization. For each Pareto-optimal design, Latin Hypercube Sampling (LHS) is used to generate virtual samples representing manufacturing variations, and their performance is rapidly predicted by the trained DL model. Finally, the Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS) is employed to systematically select the single most robust design based on both nominal performance and statistical metrics. The practical viability of the entire framework is demonstrated through the fabrication and testing of a CNC servo motor prototype. Experimental results, including both no-load and on-load tests, confirm that the manufactured motor's performance aligns with the simulation predictions, validating the proposed framework as a reliable and efficient solution for industrial applications.

키워드