Artificial Intelligence Enhanced Scaling Design Database for Electrical Machine Inverse Design

To explore the potential of generative artificial intelligence in electrical machine inverse design, this paper focus on database development as preparation for model fine-tuning and agents developing. A framework is proposed to construct the database spanning a wide range of power ratings, characterized by geometric similarity, using surface-mounted permanent magnet machines as a case study. Python-driven interactions between finite element analysis and optimization algorithms facilitate this process. Scaling and correlation factors are used as variables for finite element model construction and key performance indexes evaluation under multi-physics considerations. These factors, paired with key performance indexes, form the sample set in a single cycle. A metamodel of optimal prognosis based surrogate model is trained using 500 samples collected via Latin hypercube sampling within 23 hours, mapping factors to key performance indexes. Using this surrogate model, a genetic algorithm generates 9900 scaling designs in 10 minutes. 16 designs on the predicted pareto front were validated by finite element analysis, showing strong alignment with predictions and confirming the effectiveness of the proposed framework. Further, 4 designs were directly retrieved from the database to meet the given specifications, with No. 78, No. 3501 verified by finite element analysis showing deviations within 10%. This demonstrates a method in inverse design, eliminating the need for time-consuming fine-tuning to satisfy specifications.